Dynamic, Traffic Related Market Packages	Definition and Key Points
Broadcast Traveler Information	§ Disseminates near real time traffic information over a wide area through existing infrastructures and low-cost user equipment such as radio or cellular phones § Information flow is one way
Interactive Traveler Information	§ Responds to a user’s request with tailored information, using wide range wireless and wireline communication systems
Dynamic Route Guidance	§ Provides advanced route planning and guidance that is responsive to current conditions § Relies on a digital receiver, map databases and a variety of in-vehicle computational systems and devices
ISP-Based Route Guidance	§ Similar to Dynamic Route Guidance, but it moves the route planning function from the user device to a service provider § The user has the option of equipping their vehicle with the map databases and location determination capability
Integrated Transportation Management/ Route Guidance	§ Used by both public consumers and traffic management centers § Traffic management centers use it to optimize traffic control § Consumers benefit from advanced route planning and guidance based on current conditions
In-Vehicle Signing/Message Exchange	§ Based on communication between roadside equipment and in-vehicle devices § Roadside equipment communicates with the traffic management subsystem in order to provide traffic and travel advisory information to the in-vehicle device

Section 3 ATIS Concepts, Trends, and Directions Affecting Data Fusion

Section 1 introduced the concepts of ATIS. This section provides a deeper foundation for developing a data fusion model and guidelines by presenting the major findings from the literature review and the insights gained from discussions with representatives at selected public ATIS sites. The section concludes with key findings and implications for ATIS data fusion. The appendix of the report contains an annotated bibliography of materials that supported the development of this section.

3.1 Summary of Literature Review and Selected Interviews

ATIS services have been studied or analyzed from several perspectives, which cluster into three groupings: institutional/organizational, operational, and technological. A brief overview of these groupings is presented, with the majority of this section devoted to the more technical issues of data sources, data processing, and data/information uses necessary to support ATIS data fusion systems.

The institutional/organizational perspective represents some of the major challenges to successful ATIS data fusion, especially real-time systems. In general, the institutional issues pertain to planning and cooperation in the design and implementation of systems in order to collect and share data appropriate for traveler information. A key topic, currently under discussion at many public agencies, is the scale and scope of the public sector’s role and obligations in providing traveler information. Obviously, the purpose of a data fusion activity, the associated architecture, data collection and sharing requirements, multi-organizational/institutional issues, and similar design consideration all affect the performance and usefulness of an ATIS data fusion system. These institutional issues may be the greatest within a single agency or organization and likely involve intra-agency relationships, resource allocation, and policies that introduce new relationships and responsibilities between function groups (such as planning, design, and maintenance/operations) and support groups such as Information Technology, records management, and similar repositories or stewards of data. Institutional issues for advancing ATIS data fusion also occur across organizations in which geographic and jurisdictional boundaries create interface and data sharing challenges. Current practices to manage these interfaces include Memorandums of Agreement, performance-based specifications, and delineation of interface protocols and requirements. While these interface issues may have been achieved in such traditional areas as asset management and transit service providers, new challenges emerge when considering issues of data coverage and quality, ownership rights, and data fusion performance-based criteria. These institutional issues are discussed in this section within the context of developing a data fusion model.

Operational issues have been captured best through case studies and evaluations of ATIS. The evaluations have provided insights into a range of challenges, including institutional, technical, standards development, resource allocation, and implementation of new systems and partnerships. The operational issues are highlighted in Section 3.1.5.

The technical issues associated with developing ATIS data fusion systems can best be represented by a simplified, three-part, structured analysis. This format is widely used in this type of study to define system inputs, processes, and outputs. It also provides a simplified introduction to data fusion, in preparation for the more detailed functional representation found in Section 4.

Inputs for data fusion are from one or more sources that are collected and subsequently processed to meet specific end users’ needs or output requirements. The following figure illustrates this three-part process. The three parts of the process serve as the major discussion topics reviewed in this section.

Data Sources	Data Processing	Data and Information Uses
§ Sources and collection of data	§ Protocols and standards	§ Information services for pre-trip or en-route traveler needs
§ Data Quality	§ Data Fusion Functions	§ Users’ demand for services
		§ Dissemination of information
		§ Telematics

Figure 3-1 A Simplified Structured Analysis Model of ATIS Data Fusion

Many of the institutional, operational, and technical perspectives overlap in this simplified model since they contribute in one or more ways to the overall functions and performance of an ATIS data fusion system. The separation shown in the figure, however, does allow a convenient means of highlighting the major data-centric findings that contribute to the opportunities and challenges for ATIS data fusion.

The following discussions focus primarily on real-time freeway and major arterial systems. However, archived data, data from speed studies, and other traffic or transit studies also constitute valuable sources of ATIS data. The challenges of using these data sources will be ones of compatible frames of reference (geospatial and temporal), data verification, and data formats. These current issues and practices are discussed in the following subsections and addressed in Sections 4 and 5 of this report.

3.1.1 ATIS Data Sources

Two topics dominate the literature associated with ATIS data sources: (1) a description and discussion of the many sources of data and information that can contribute to ATIS services, and (2) input data quality.

3.1.1.1 Sources and Collection of Data

The largest source of ATIS data is the loop detector system, usually owned, operated, and maintained by the public highway agency. Loop detection data requires some level of localized signal processing after which the modified signal is communicated to a more centralized location, usually based on a polling request. Loop detector data can be processed to estimate, among other things, speeds, occupancy, and congestion. Typically loop detectors are placed approximately one-half mile apart on highways in metropolitan areas and may be placed around interchanges or intersections to collect data about traffic flows[5]. Major arterials may have loop detectors, but usually only at key intersections or near highway connections. Key determinants in placing detectors on arterials include consideration of studies of traffic counts, safety reports, availability to communication networks, and usability of the collected data. As data collection coverage needs increase, arterials are becoming more instrumented.

Loop detection reliability[h] has been estimated to be between 10 and 50 percent, depending on agency maintenance practices, time of the year, and age/type of the loop detector. Some loop detectors are more than 30 years old, which affects performance, accuracy, and the type of information they are able to collect. Communication links between field collection points and a centralized location may be affected by a variety of factors including telecommunications engineering (speed, bandwidth, communications protocol, polling frequency, transmission errors and reprocessing of data streams, noise, etc.) and maintenance issues (cable cuts, power outages, etc.).

Once collected, field data is usually stored on a large capacity, non-proprietary database system. These database systems may collect as much as 2GB of daily information, depending on the size of the network, the sampling rates, and data compression, if used. ATIS systems make use of real-time as well as historic data. Consequently, data management, data messages, and data access issues can affect overall ATIS system performance. Efficient and cost effective storage of video data is still an evolving field of study[6].

Loop detectors only measure aggregate traffic at particular freeway locations, rather than the movement of individual vehicles in a specific space. Consequently, ATIS route guidance services are severely hampered by the lack of information about trip distributions. To augment loop data, additional sources of data and information are being used. Video technology has been developed and applied for more than 30 years as a substitute for inductive loop detectors. Within the past 3-5 years, CCTVs have migrated from analogue to digital formats, resulting in greater resolution and lower costs. Three major developments have enhanced the use of CCTV for ATIS services. First, the digital conversion allows for more efficient signal processing and field-to-center transmission, including the use of compression methods such as the de facto MPEG-2 standard. Higher resolution images could enable such activities as improved incident management or vehicle tracking at sensor locations, thus contributing to discerning trip distribution. Second, digital video recorders are being produced in sufficient quantity to make them cost-effective means of storing video images, especially when coupled to video motion detection methods that do not engage the recorder when no significant activity is detected, thereby saving storage space. Third, low cost, efficient transmission protocols have emerged in metropolitan areas to enable more reliable and wider network coverage. These protocols are based on such technologies as enhanced fiber optic networks (Dense Wave Division Multiplexing- DWDM), Digital Subscriber Line (DSL) protocols, Multi Protocol Labeling System (MPLS), and others.

The effectiveness of CCTV for ATIS use is a function of image resolution, communications bandwidth, uses of the image (human interpretation versus pattern recognition methods), and overall life cycle benefit-cost. Video cameras are used to detect and verify roadway or network conditions and confirm incidents within the range and capabilities of the equipment and telecommunications system. Video information is centrally collected and manually scanned for unusual traffic patterns. Some traffic management centers have the capability for simultaneously displaying more than 50 field locations, requiring additional personnel or less frequent reviews of the video images and assessment of the traffic networks. Camera network coverage and deployment is typically constrained by the cost effective access to higher-bandwidth communication links. Moreover, the quality of the video does not readily lend itself to pattern recognition or feature extraction (e.g., license plate) processing due to low resolution and frame refresh specifications.

Another source of roadway information comes from travelers who report events or incidents. These reports provide a varying range of usable data about the event. Many times drivers are unable to accurately locate where the event occurred, although GPS chips in cellular devices may allow for greater accuracy. Moreover, this information is conveyed to a human operator, who must be skilled at eliciting the information needed to verify and characterize the event in order to provide sufficient traveler information. However, with the “human-in-the-loop,” there will be restrictions on the number of calls that can be processed, regardless of the operators’ abilities.

Another source of traveler information comes from vehicular “tags” (RFID devices) that can be read by roadside readers on public roadways for the purpose of charging a user fee (toll tag), identifying vehicles (CVO registration or transit vehicles) or verifying credentials (WIM specifications). This information is usually available only to the public agency responsible for collecting a fee or processing traveler information. The information would be valuable for developing an improved understanding of trip distributions, but currently there are no institutional arrangements that allow for sharing this user information with ATIS service providers, public or private. Other sources of traveler information come from GPS-based systems, including certain types of Automatic Vehicle Location (AVL) systems employed in certain transit systems and GPS-enabled probe vehicles.

Once the data is collected, it is usually stored in a database waiting further processing.

3.1.1.2 ATIS Data Quality And Performance Issues

In general, ATIS data quality can be defined as fitness for use by information consumers, public or private. Determining the fitness for use is a complex issue to assess since the data quality can originate in the data source, data processing, and application or use of the data. At any point in this process, data quality can be enhanced or diminished.

Several sources of poor ATIS performance have been noted in studies3,10,14, including:

· Scope and scale of data coverage (geographic and temporal)

· Software errors (development and maintenance phases)

· Hardware or facility failures (including security breaches)

· Poor input data quality

Data coverage is one of the more important sources of poor ATIS performance primarily because sensors and other sources of data collection do not adequately cover the network to be associated and supported by data fusion. Moreover, the roadway networks that are covered tend to be freeways rather than major arterials and key intersections. Transit systems coverage is similarly hampered by the availability of temporal and geospatial vehicle information. Enhanced software methods for improved code development are being applied in the ATIS field, including the use of structured design methods, application of costing/performance models, and use of more rigorous contract specifications to ensure performance, standards, and warranty issues are addressed. Hardware failure, especially at the distributed sensor level (such as inductive loop detectors), is an individual agency issue. Some agencies have indicated that as much as 30 percent (on average) of their network of loop detectors may be unavailable or provide inaccurate data on a daily basis. Failures at centralized facility or operation centers are less of a problem since private-sector recovery practices have been applied in public agencies. These practices include facility designs with disaster recovery features and hardware configurations to provide frequent backup of computer and communication systems (RAID, frequent imaging at locations unaffected by local disturbances, communication system redundancy, etc.). Field facilities usually employ environmentally hardened equipment and power supply configurations to minimize outages. When field failures do occur, automated alert systems usually become activated or else central facility operators detect a malfunction or outage. Poor input data quality has been identified as the most common source of information system failure. While there are no estimates available from the ATIS community, business organizations that have reported data quality issues estimate the data error rates ranging from one-half to 30 percent[7].

The literature contains numerous ways and specific techniques to improve data quality. These techniques can be classified into two broad categories:

· Finding and fixing data errors (clean-up)

· Preventing errors

The ATIS service providers interviewed for this project indicated they only focused on the first category, although many would like to have the more systemic approach of preventing errors. After determining that poor input data quality was the source of concern, their typical response to poor input data quality was a three-part process:

(1) determine the upstream sources of the erred data (sensors, communication links, etc.) and resolve as appropriate;

(2) conduct large or small database clean-ups as part of everyday maintenance operations (correcting obvious errors such as outliers, false alarms, no signal, no value, etc.); and

(3) deal with the impacts of the erred data, such as manually correcting ATIS information using historical estimates (if appropriate), posting notifications of information unavailability, searching for the erred data source (if not already found, isolated, or qualified).

The effectiveness of this three-part process has not been assessed systematically in the ATIS community. However, experience from other organizations offers some insights. The sensor or communications link data quality issues are becoming relatively easier to resolve as improved equipment with self-diagnosing methods is deployed. In addition, maintenance and operations functions in public agencies are becoming more tightly coupled as performance-based management systems are being implemented. These organizational process improvements allow ATIS information providers to be more aware of operational issues associated with network and equipment performance. The database clean-up strategy works in the short-run, but is less effective in the long-term. While certain automated tools exist to correct the obvious errors, more errors are usually created due to newly calculated data elements based on uncorrected, erred data. Depending on the frequency of the use of the data (e.g., monthly reports on ATIS usage versus hourly indicators of signal system timing), the data may become permeated with errors. This leads to a never-ending, time-consuming cycle of periodic clean-ups, at increasing expense. Short-term database clean-ups are done on time-sensitive or system-sensitive data, but also face the same long-term issues. Moreover, this database clean-up approach does not provide a long-term, scalable answer[8].

3.1.2 ATIS Data Processing

ATIS data processing is based primarily on systems and applications developed in other disciplines or industries. One of the most important issues in ITS, transportation, and traffic communities is the lack of widely recognized and accepted communications and data standards. As a consequence, challenges of interoperability and interchangeability among agencies, across jurisdictions, and among users have hampered data sharing[i], data processing, and more widespread use of ATIS services. Increased standards and protocols definition and adoption has been discussed extensively in the literature and seen by some as an evolutionary means to help reduce institutional and organizational barriers in planning and implementing data fusion systems. After data and communication standards, data processing methods and approaches emerges as a key topic in the advancement of data fusion systems. Each of these topics is reviewed in the following subsections.

3.1.2.1 Standards and Protocols

Data about travel conditions are the foundation for ATIS services. The data is sensed, collected, transmitted, and processed by multiple agencies at different locations using different types of equipment. The resulting information is used to support traveler information services, but also traffic management operations, maintenance activities, emergency response, as well as agency planning and budgeting functions. Typical systems have several integration and interoperability issues such as:

· Different types of devices unable to operate on the same communications channel

· Software system updates as new devices are added

· Proprietary protocols or device configuration information

· The inability to effectively exchange information among public and private agencies supporting ATIS and other transportation functions

To ensure these needs are met and all subsystems are integrated and interoperable, as much as possible, standards are being developed. There are three primary categories of ATIS standards being researched and developed: (1) communication standards; (2) data dictionary elements (objects); and (3) message sets. The purpose of the standards development process is to:

· Increase design flexibility and choices for operating transportation agencies

· Enhance interoperability and coordination

· Remove proprietary barriers

· Allow different types of devices and enabling software from different vendors to be mixed within the same system at minimal integration life cycle costs

For the most part, agencies and industry participants have embraced this approach and are contributing to its success.

The overall standards process is a multi-part endeavor:

(1) Identifying the standards to be defined

(2) Defining the standards

(3) Testing/verifying the standards

(4) Maintaining the standards

(5) Training in the standards

(6) Communicating/outreach on the standards

Most of the effort to date has focused on the needs and definition of standards. More than 100 standards have been identified and developed to varying degrees of completion. For traveler information, a variety of standards have been defined and are under development, including the following key elements:

· Vehicle Location Referencing Standards

· In-Vehicle Message Priority Sets

· The message set for Traveler Information Systems

· Data Dictionary Standards

· The Traffic Management Data Dictionary

· NTCIP-Compliant Dynamic Message Signs

· Many others [j]

Comprehensive testing of the standards is a process that is just beginning.

3.1.2.1.1 Communication Standards

The National Transportation Communications for Intelligent Transportation System (ITS) Protocol (NTCIP) is a family of standards that provides both the rules for communicating (called protocols) and the vocabulary (called objects) necessary to allow electronic transportation control equipment from different manufacturers to operate with each other as a systemc. These are open, industry-based standards that make it possible for ATIS and other ITS services from multiple vendors to exchange information using a common communications interface. The NTCIP is a five-layer model developed, in part, on the Open Systems Interconnection (OSI) 7-layer concept[k]. There are a variety of NTCIP standards pertaining to the level and interface within or across levels. The NTCIP is the first set of standards for the transportation industry that allows transportation management and control systems to be built using a "mix and match" approach with equipment from different manufacturers. Therefore, NTCIP standards reduce the need for reliance on specific equipment vendors, customized one-of-a-kind software, and costly interface management. NTCIP is a joint product of the National Electronics Manufacturers Association (NEMA), the American Association of State Highway and Transportation Officials (AASHTO), and the Institute of Transportation Engineers (ITE).

The NTCIP defines two major types of communications, both of which are important to ATIS:

Communications between a management center and field devices (Center-to-Field – C2F)
Communications between two or more management centers (Center-to-Center – C2C) as well as units within the same agency[l]

C2F communications may involve commands to field devices or the receipt of data from field devices, such as polled roadway loop sensors or CCTV images. C2C communications enables information exchange about signal status, incident/event notification, image sharing (CCTVs), and regional traffic control (major event traffic diversion control and implementation).

3.1.2.1.2 Data Object and Message Set Standards

Of particular concern to ATIS are the information-level and applications-level standards that address center-to center communications, since traveler information usually does not reside at one center (or database) location (e.g., roadway, transit, weather). A variety of standardized data elements (objects) and message sets supporting ATIS have been developed. Message sets are common terms and definitions for ATIS data used to disseminate ATIS information. In a simple analogy, message sets are the sentences whereas the data elements are the individual words. The standards supporting ATIS functions and services include:

ATIS Data Dictionary, which provides the definition and syntax of individual data elements that make up the specific message content of a message. SAE J2353, Advanced Traveler Information Systems (ATIS) Data Dictionary, defines the data elements for ATIS messages and was approved in August 2000.
ATIS Message Sets, which provides the messages that are exchanged among information providers, traffic management centers, and other ITS centers. SAE J2354, Advanced Traveler Information Systems (ATIS) Message Sets was approved in October 2000.
Traffic Management Data Dictionary developed by ITE, which supports C2C information exchanges, is under review and comment.
Message Sets for External Traffic Management Center Communications (MS/ETMCC).
Incident Management standards, which are being co-developed by NEMA and IEEE, are being developed.
Roadway Weather Information System (RWIS) data and message sets are being developed for siting and calibration needs, as well as communications[9].

At the applications level, six protocols have been proposed and standardized by NTCIP. These include Common Object Request Broker Architecture (CORBA[m]), DATa Exchange in Abstract Syntax Notation One (DATEX-ASN.1), Simple Network Management Protocol (SNMP), Simple Transportation Management Protocol (STMP), File Transfer Protocol (FTP), and Trivial File Transfer Protocol (TFTP). CORBA and DATEX-ASN.1[n] are suitable for C2C communications whereas SNMP and STMP are more appropriate for C2F. FTP and TFTP can be used to retrieve server files when a client receives a file transfer request. With the exception of STMP and DATEX-ASN.1, all application protocols exist and are widely accepted industrial standards for the application layer.

3.1.2.1.3 Metadata Standards

In an increasingly distributed computing and communication environment, it will be necessary to define content description standards or metadata standards of complex, multi-layered, time-depending data streams. A metadata model specifies an application's object, structure and content and conveys information about the application's elements thereby enabling reasoning about the data and the metadata. A variety of approaches are emerging on metadata definitions including the Resource Description Framework (RDF) Schema. Extensible Markup Language (XML)[o], Document-type Definitions (DTDs), Document Content Description (DCD), MPEG-7, and Schema for Object-Oriented XML (SOX). The ATIS developments in Seattle have been employing self-describing data (SDD) techniques since the mid-1990s[10].

3.1.2.2 Data Fusion Methods and Approaches

The data fusion field has been in existence for about 20 years. Consequently, some of the methods and techniques applied in data fusion work draw heavily from other disciplines and fields of study and application. The arena of data fusion is best captured in the following figure. As illustrated in Figure 3-2, a number of disciplines, areas of study, and techniques contribute to the data fusion.

Figure 3-2 Data Fusion Draws From And Contributes To A Number of Overlapping Disciplines

Figure 3-2 Data Fusion Draws From And Contributes To A Number of Overlapping Disciplines[11]

Data mining is the nontrivial discovery of meaningful new correlations, patterns, and trends and the extraction of implicit, previously unknown, and potentially useful information from large amounts of data. By applying sophisticated software tools, an analyst is able to infer rules from among data objects that can be used to predict future states or guide decision-making. Data mining methods represent one of several techniques for developing some of the parameters needed by more advanced data fusion algorithms, such as knowledge-based systems or intelligent agent functions, which are discussed below.
Estimation refers to the use of methods to infer information or define parameters about a general population based on a limited set of observations about the population. Estimation methods usually are based on principles of sampling theory and include metrics for the accuracy of estimation.
Sensor Management refers to the range of activities to ensure sensor data is formatted and processed in a timely and accurate manner as needed by the data fusion subsystems, as well as possibly controlling the sensor operations[p] and adjusting the data processing in order to improve estimation and prediction of selected objects.
Correlation refers to the degree of relationship among two or more variables.
Tracking refers to the temporal and geospatial location of objects. The tracking of objects involves a number of factors including uncertainty in position location, management of multiple measurements for the same object, database organization, and scalability.

ATIS data fusion development can be categorized between data-centric activities and model-centric activities. Data-centric systems typically analyze large pools of data found in the host agencies databases and may be augmented by external data sources. The analysis and fusing of the data is typically a well-structured problem and involves the timely collection, processing, storing of information at proper locations (e.g., data warehouses), and disseminating (“pushing”) the information based on the ATIS system design or subscriber services. Data elements can be part of an object-oriented or relational database management system. Model-centric systems are heavily based on algorithms that employ some form of reasoning to assess a current situation and to provide forecasts or estimates. A model-centric system may draw on the data-centric information to make informed estimates of patterns, trends, and “what if” types of situations. For ATIS applications, these estimates may include a variety of topics, including a forecast of network clearance times given the current depiction of an event or the estimated changes in historic mean speeds on highways when rain is occurring at a certain pace and time of day. Figure 3-3 illustrates the general distinction between data-centric and model-centric ATIS data fusion functions. Current ATIS systems tend to be data-centric.

The technology and methodologies used in data fusion are rapidly evolving in a variety of fields, including military applications, medical diagnosis, and industrial process controls/operations. With distributed database systems and the rapid generation of data that may support ATIS functions, renewed interest in data fusion methods has emerged. Current research is focusing on the development of new or improved algorithms and the assembly of techniques into an overarching framework and architecture that addresses a data fusion project’s needs. An overarching framework also assists in comparing findings from related fields (e.g., correlation and tracking analysis, while extending studies into unresolved areas).

Figure 3-3 Data Centric and Model-Centric ATIS Data Fusion Activities

3.1.3 ATIS Information Uses

The uses of ATIS information have received the most attention and discussion. Early studies focused on the range of information services that may appeal to users for both pre-trip and en-route circumstances. Later attention focused on the various means for distributing the information. As experience with ATIS deployments continued, subsequent studies and reports examined the various business models and users’ demand for ATIS services. Convergence in communication systems and in-vehicle technologies brought renewed focus to the telematics field, which offers insights into IVN performance, system design and implementation, as well as estimates of market demand and price points for various services.

3.1.3.1 Information Services for Pre-Trip Needs

Most pre-trip information systems are demonstrations of individual technologies or configurations of technologies that provide pre-trip information to a wide-range of users, based on their requests and actions. Given the relatively static nature of pre-trip information, agencies or service providers are able to organize the traveler information using Commercial-Of-The-Shelf (COTS) software (with minor modifications) and distribution methods, e.g., an agency’s web page, e-mail address, telephonic voice message tree, and PDAs. The pre-trip information is assembled into useful information services (“packages”) based on expert judgments, feedback from users, and the evolutionary refinement of the business models, e.g., adding desired features based on budget cycles, service demands, the cost/value ratio of implementation. The primary challenges for pre-trip ATIS are the need to keep timely and accurate information in databases (especially timely information across jurisdictions), the operation and maintenance of the dissemination technologies (e.g., modem-connections to devices, updates to voice messages and/or recognition software to offer schedule information, web-page updates). During the initial trials of deploying traveler information websites, some agencies were not able to handle the volume of requests and consequently, potential users were turned away due to unacceptable delays in response time. These issues have been largely overcome with enhanced network sizing, planning, rapidly declining equipment costs, and improved processing speeds for web servers.

3.1.3.2 Information Services for En-Route Needs

En-route information services typically provide travelers with a variety of real-time traffic conditions, incidents, construction, weather, and transit schedules/operations.

Currently, many state agencies collect and process data to yield some form of traffic conditions, usually average speeds or camera images associated with selected highway segments. Typically this information is centrally assembled and analyzed at a data warehouse or traffic management center, where co-location with law enforcement and/or agency maintenance personnel may also enhance the information regarding the scope, location, accuracy, and timeliness of the highway or network conditions. The co-location of these supporting services usually allows for more rapid detecting, tracking, and promulgating of time-sensitive information about incidents, roadway construction, or closures to ATIS users[q]. Some of the primary issues for en-route information services are the extent of the network coverage (sensor spacing/location as well as type of roadway -- highways, but not usually adjacent or key arterials), accuracy and availability of the data within and across jurisdictions, and timeliness of the data[12].

Value-added information about roadway conditions may also be generated by a third-party Information Service Provider (ISP) who further refines and augments basic network conditions by providing enhanced estimates of travel times or delays, suggested routes or navigation information based on forecasted roadway conditions, pre-trip advice/planning based on current network conditions, and other services. Third-party information providers may use additional data collection methods, such as aerial surveillance and probe vehicles[r] or capture information from travelers, passively through ancillary location detection via cellular phone calls[s] or actively through direct reports of events to a TMC, emergency number, or local radio station. This enhanced information may be provided for little or no cost through commercial radio broadcasts or on a fee-for-service basis.

Turn-by-turn route guidance, usually with voice or tone prompts, is another form of ATIS that has received generally favorable support and use. These systems are undergoing additional enhancements through the use of better display maps, faster in-vehicle processing and response time, GPS equipment offering better accuracy and signal reception, and experimental voice-activated commands to meet “hands-free” driver requirements. These product improvements are driven by a combination of cost-effective technological enhancements drawn from other industries and favorable value/cost ratios based on user feedback and preferences. The rental car industry is one of the primary providers of these ATIS services, starting with the initial Travtek project[13].

En-route transit services are beginning to offer expanded information on transfers and bus/rail connections, arrival and departure times at designated stops, and space availability at “park-n-ride” lots. These enhancements are typically the result of combining market proven technologies into a new “package” of ATIS services. For example, with the recent California certification of BART’s automatic vehicle control system, even greater precision in the geo-location of vehicles through dedicated wireless networks will be possible, enhancing general operations and specifically the accuracy, reliability, data quality, and timeliness of arrival and departure information.

3.1.3.3 Users’ Demand for ATIS

Almost all ATIS users appear to be employed commuters with the highest usage during peak commuting hours [14]. Users generally react favorably to having the choice of receiving accurate, real-time traveler information. In general, four factors influence the level of use or demand for ATIS services:

Regional traffic context. The primary users for ATIS come from highly congested regions with limited build-out options, constrained route possibilities, and frequent traffic events due to incidents, weather, unexpected roadways repairs, etc. Accurate, regionally oriented, timely message formulations for ATIS users are critical.
Quality of the ATIS service. Users are interested in the accuracy, reliability, timeliness, portability, and value/cost ratio of the ATIS information. These factors influence the level of confidence and how frequently an ATIS service would be consulted.
Individual travel or trip characteristics. Trip length, particular routes or choices, departure time flexibility, and alternative mode choices influence the demand for ATIS.
Characteristics of the traveler. Factors such as user awareness of available ATIS services, potential use patterns, responses to services provided, human-machine interface design, and valuation influence to varying degrees the demand for and types of services needed or requested.

Focused research on ATIS users’ needs has been building for the past 8-10 years and is still on going. While these four factors indicate the extent to which users will consume ATIS services, much more work is needed to understand such quantitative and qualitative factors as desired products/services, users’ needs for timeliness, accuracy or reliability, and the user’s “willingness to pay” or price breakpoints. These users’ needs, in turn, help system developers establish technical requirements and their relative priorities in the ATIS system and among subsystems.

3.1.3.4 Dissemination of ATIS Information

The primary means of disseminating traveler information has been through audio (telephone, radio broadcasts, etc.) or video means (CCTV, CATV, broadcast TV, web-based services, etc.). The specific means and methods have largely been imported from developments in other industries.

Websites have been one of the most common means used by ATIS service providers of disseminating traveler information. Most of the website development is done under agency direction using their preferred web development standards. Consequently, there are no standard formats, metadata, or descriptors established for the websites[t]. While languages such as HTML and XML provide for document definition techniques, the implementation of the websites do not necessarily follow a standard format or set of techniques. Consequently, web-based data mining and data parsing is usually conducted on a case-by-case basis. As webpage formats change, adjustment in parsing, mining, and capturing techniques are usually required. Recent advances have employed machine learning approaches to effectively parse and output in XML formats the content of a web site[15].

3.1.3.5 Telematics

Telematics refers to automotive-grade electronics and communications systems to make vehicles operate smarter, safer, simpler and more synchronized with their environment. Telematics can assist with location-based services, entertainment, navigation, concierge services, and emergency needs. Telematic components include a variety of technologies and processes, including GPS technology, radio frequency communication subsystems, embedded computing to monitor vehicle performance (mechanical, electrical, driving behavior, road conditions), on-board devices for display (flat panel displays, audio systems) and capturing information (hands-free operations, voice-activated commands). The telematics industry offers a variety of services, including identification/location, navigation, entertainment, personal safety and security. A subset of these services may use ATIS data sources and information.

Telematics is an evolving industry. There are an estimated two million vehicles with some level of ATIS telematics capability installed, however these estimates are difficult to verify since automobile manufacturers rarely publish information about their telematic sales and after-sale service agreements[16]. Companies are experimenting with a variety of business models to determine consumer preferences. Technically, telematics evolution will progress based on the establishment of industry standards and the use of cost-effective hardware and software components appropriate for automobile/truck/bus use[17]. Some standards for in-vehicle communications protocols have been established (e.g., in-vehicle data bus (IDB) standards), however others are still working through the standards development organizations, such as common incident message sets and roadside-to-vehicle protocols.

3.1.4 Legal/Institutional Issues

Many agencies use their own data to provide basic traveler information (e.g., average roadway speeds, roadway images for local broadcast television, transit schedules). The initial legal concerns about tort liability (distributed information that is false, inaccurate, or unreliable resulting in an accident or damages) have been addressed by developing disclaimers of liability and the use of warranty conditions. Liability also has been addressed through the use of third-party state universities who collect and distribute the information under research-based agreements, which minimize the potential for claims.

Recent research work has focused on data sharing, specifically how the public and private sectors deal with data ownership and sharing, and examines policies aimed at facilitating data sharing and ultimately improving the quality and quantity of information that reaches travelers [18]. Even though their motives are typically different, public agencies and private sector organizations are both active participants in the use of traveler information as a transportation management tool. A major finding was that agencies that have data to share protect their interests by placing restrictions on access to data, but firms generally do not find these conditions to be onerous. Almost all agencies provide information directly to the public with VMS, HAR, kiosks, and interactive voice response telephones. Although agency data are a fundamental source, private providers generally need to enhance public data before they are marketable. The most common types of augmented information include traffic and road conditions, incident information, and planned construction information. Transit data are generally less useful to private providers, and only a third of them report transit delay information. When new equipment and operating expenses were required, the primary beneficiary incurred the expenses[u]. Finally, two or more conditions on data access were observed, the most frequent condition being an acknowledgement of the agency as the source of the data when distributed to the public (e.g., a logo on an image or a statement such as “this information brought to you courtesy of ‘agency’ …”).

3.1.5 Site-Specific Evaluations

A number of evaluations of ATIS services have been conducted, either as stand-alone systems or as part of a larger evaluation effort. Two major sources of evaluation are the Field Operational Tests (FOT) evaluations of ATIS and the Metropolitan Model Deployment Initiatives (MMDI) assessments. Verifiable private sector assessments were not available. Three representative ATIS studies are summarized since they provide insights into operational issues that affect ATIS data fusion system design and operation.

An FOT cross-cutting evaluation of ATIS was conducted in 1998[19]. The primary conclusion was the users value traveler information, but usage will depend on a variety of factors such as awareness of service, information reliability, timeliness, accuracy, and user cost. At the time of the FOT study, key issues concerned legal liability, improved business models for partnering, expectations for agency-partner performance, quality in a quasi-private business enterprise, and market research into users’ preferences. Some of these concerns have been investigated further or since resolved.

The MMDI evaluations were multi-purpose evaluations to assess the deployment and efficacy of ITS. Certain studies focused on ATIS, such as those in Seattle and San Antonio. In Seattle, the evaluation focused on traveler use of the Washington State Department of Transportation (WSDOT) traveler information web site in Seattle and the greater Puget Sound area[20]. The objective of this web-use analysis was to better understand preferences for traveler information, the usage levels at the MMDI site areas for information provided through the Internet, the potential for reaching the traveling public via the Internet, and the patterning of use of traveler information. Findings indicated the WSDOT traffic conditions web page is very popular with travelers in the Seattle area and is one of the most heavily used traffic information web sites in the nation. Unusually congested traffic, caused by severe weather or incidents, prompts significant “spikes” in usage of the web site, indicating substantial latent demand for real-time traffic information. The CCTV image pages were heavily frequented. The daily patterns of use clearly indicate heaviest usage during the afternoon commute peak period.

In San Antonio, the MMDI evaluation of the Transguide system reported the most effective stand-alone ATIS implementation was the incident management component[21]. The ATIS system utilized kiosks and a web site as the primary means for disseminating information. The kiosks had several functional problems and location/placement concerns, thus they were less unlikely to be used by travelers. The web site evaluation indicated substantial use of the site, especially for the real-time weather and traffic/roadway conditions. Overall, the pre-trip and real-time ATIS services indicated they could help reduce delay, crash risk, and fuel consumption.

Discussion with operators and planners at six public ATIS sites were conducted. The sites included Seattle, San Francisco, Los Angeles, Houston, the I-95 Corridor Coalition (Virginia/Maryland representatives), Hampton Roads—Smart Travel Center/I-81 area, and Transcom.

General observations from these discussions include the following:

The primary sources of real-time data are the loop detection system, CCTVs, video imaging detectors, tollgates, reports from law enforcement or agency patrols, weather-sensing stations, other operations centers (transit, bridge-tunnel authorities) or travelers. In general, data is collected in the following categories: traffic (speed, volume, congestion, lane occupancy, road conditions, weather, parking); incidents and emergency management (accidents, special event, road closure, construction schedule); transit (metro schedule, fares, intermodal connection); and other more customized special services including ride-sharing, location and navigation information.

Data is almost always gathered at a central location, processed, and then distributed. Any data fusion activity is centrally based.

Data fusion activities are either manually performed by layering data over maps or combining information through human judgments (e.g., a human operator verifies a reported incident and defines subsequent actions). The resulting assessment and actions are usually monitored or confirmed by an operations “chief” and posted for dissemination as a verbal report, web-site update, or DMS display. Only when using a self-contained system, such as with a GIS or CAD system, was a more automated data fusion approach possible, primarily through relatively simple geo-spatial and temporal alignment algorithms. Most GIS and CAD systems manipulate proprietary data sets and so interoperability and exchangeability of data is complicated[22]. One agency has five different GIS systems used throughout the agency, each of which would be valuable to ATIS services. However, automated cross-referencing in a cost effective and timely manner across the five systems is not possible.

The most frequently cited reason for insufficient data quality is inadequate geographical coverage. This arises mainly from incomplete data collection in metropolitan areas with multiple jurisdictions with incompatible data sources, ITS, or ATIS system within each area.

The extent of archived data to be used, especially for travel time estimates, has not been well defined. Both real-time and archived data are important for travel time estimation. Techniques for using this information have been left to research universities or other interested individuals. While most agencies are not out of data storage space, the cost of archiving may eventually impose limitations on archival and timely retrieval capabilities.

An increasing number of ATIS services have incorporated GPS functions. There has been an emerging common practice of using Location References Management System (LRMS) to incorporate GPS functions into ATIS services. In terms of incorporating GIS functions, achieving interoperability (interchangeability) is more complicated, because there are already many GIS products in various formats and existing practices.

Data quality is assessed primarily by informal or heuristic judgment methods. The extent of the assessment is in reaction to the identification of outliers or obvious errors. Some quality check functions are embedded in the process (hardware/software), but there is still no common approach or organizational discipline to establish data quality rules. Data collected through some loop systems did provide metadata, which offer the means of assessing rudimentary single loop performance (e.g., recent calibration, false signal alarms, equipment self-reports).

Data is shared through third-party partners or lead agencies that have the consent of the other partners. No fee is usually charged, but acknowledgement of the source is required.

Agencies are migrating to the use of the NTCIP and associated standards, but will likely phase in their use based on industry acceptance and agency confidence in their usefulness and longevity.

Many 511 systems are beginning operations. Insufficient data was available to make an informed systemic assessment.

As mentioned earlier, these sites were selected primarily because of their advanced planning and operation of ATIS services. Not all observations pertain to all sites and so it would be inappropriate to attempt to generalize these findings.

3.2 Key Findings and Implications for ATIS Data Fusion Development

There is a general awareness in the ATIS community of data fusion function and purpose. The awareness is due to the promulgation of National ITS architecture materials and specific ATIS presentations. Within the ATIS community there are multiple interpretations of data fusion and what can be achieved. One perspective is a relatively simplistic view and follows the data-centric model, e.g., primarily data alignment processing through templates or common referencing methods, with limited data analysis. Another perspective is a more sophisticated, model-centric view of the opportunities to combine data from a wide-range of data sources and then apply association and estimation techniques to meet a large set of ATIS user needs.

Data fusion is used primarily at public agencies to perform spatial or temporal alignment of input data. The data alignment and transformation issues can be addressed through proper system design and application of proven techniques. Third-party ISPs typically augment the public data, provide some form of value-added service, and offer it through a variety of mediums to fee-paying ATIS subscribers. The value-added services include greater network coverage, more frequent updates of network conditions, cross-platform compatibility and functionality for sending or receiving traveler information, and enhanced cross-referencing of complementary data/information sources, such as better maps and weather information. Some public ATIS service providers are interested in the development of data fusion methods, but do not perceive them as critical for their role in ATIS service delivery at this time. Moreover, public agencies would prefer a system-level proof of concept from another agency (reliable, durable, accurate, value/cost assessment) before embarking on more extensive data fusion projects.

Some of the challenges for ATIS data fusion development include:

For ISPs, uncertainty in the ATIS business models and potential benefits, including:

Data access issues, which three to five years ago were issues between public agencies and third parties, have been largely resolved.
Sensor coverage (primarily inductive loops, CCTVs) and reliability does not allow for more sophisticated travel time predictions or network status, which potential customers may desire;

Certain algorithms require probabilistic or other sophisticated parameters or estimates, which would need to be calculated or estimated by humans. The alternative is to keep a “person-in-the-loop” at key decision points in the overall processing of traveler information;
Uncertainty in the subsystems and system-wide performance based on the emerging standards–NTCIP, data objects and message set development status (ATIS, TMDD, RWIS, etc.);
Data fusion techniques exist but need adaptation to the ATIS field and sharing of performance across with the community;
Model-centric data fusion relies on higher-order problem solving forms (reasoning with uncertainty types of problems) which are highly context-dependent, potentially unstable, and not easily generalizable; and
ATIS systems are not typically developed as stand-alone systems and therefore co-exist and are influenced by other system architecture design and goals such as a freeway management system. Consequently, ATIS issues involve the relative priorities placed by agencies on ATIS goals vis-à-vis other ITS subsystem goals, the need for real-time processing capabilities to meet ATIS outcomes, the extent of the network sensor coverage required to achieve ATIS performance goals versus other ITS subsystem goals, and the means of defining, calibrating, and improving data quality in a systematic manner.

3.3 Section 3 References

Section 4 Data Fusion Framework for ATIS Analysis and Implementation

The purpose of establishing a data fusion framework is to provide a functional model for use by a diverse set of individuals or communities interested in ATIS. Working from a common framework, interested individuals would be more inclined to coordinate and collaborate across disciplines and expertise. A functional ATIS data fusion model also facilitates user understanding and communication, permits comparison and integration across disciplines, promotes expandability, modularity, and reusability, and offers cost-effective systems engineering and systems development insights.

This section presents an adaptation of a Defense Department data fusion model to the ATIS context. The first part defines the key functions and multi-level (or phased) features of the model as well as methods and processing techniques appropriate to each level[v]. The discussion presents a combination of data-centric and model-centric methods for advancing ATIS data fusion. The second part presents a review of possible fusion architectures with an emphasis on ATIS communication configurations and qualitative performance tradeoffs. The final section focuses on data quality assessment and management, although data quality implications are noted throughout this section.

4.1 Key Functions To Be Performed In An ATIS Data Fusion System

Section 3 introduced a simplified, three-part model to enable a general review of ATIS studies and discuss the supporting role of data fusion. That model depicted data sources, data processing, and data/information uses (outputs) as the three key functions. To advance the applicability of ATIS data fusion, an expanded model of these functions is desirable. It was noted that ATIS data fusion is evolving from a data-centric practice to more model-centric features in order to meet users’ needs. The model-centric features will require a broad range of data fusion functions, which when properly combined, can help meet the users’ needs. Moreover, new and emerging technologies will provide alternative and unforeseen means of supporting data fusion. A more generalized model can assist in incorporating these innovations. Finally, an industry-defined set of data fusion functions allows planners and developers to draw on the experience and ideas from affiliated data fusion applications.

The major data fusion functions, in support of ATIS, can be established as²:

Alignment of sensor data to a common reference of time and space (or reference frame).
Association between measurements or observable data collected from different sensors and tracks[w] to determine candidates for correlation processes.
Correlation of tracks and the measurement data to improve detection, classification, and tracking of objects of interest. Objects of interest may include vehicles, temperature, incident images, transit vehicle locations, and similar objects which, when combined, will provide a more complete representation of the transportation system.
Classification of the track data sets to determine object type, situation/network description, and impact assessments.
Estimation or prediction of an object's future position or representation.
Regulation and synchronization of data processing functions to manage the feedback of threshold levels, perform periodic self-diagnostics, coordinate integrated timing, and monitor data communications.

While listed individually, these key functions work in a concerted manner to support the goals and purpose of a data fusion system. For example, regulation and synchronization of data processing functions is a continuous, crosscutting activity. Other functions pertain only to specific data collection or object identification requirements. A functional model can be used to place these key functions into a hierarchical arrangement that support the basic purpose of data fusion, namely to combine data to estimate or predict the state of some aspect of the surface transportation world.

4.2 Data Fusion Model Applicable to ATIS

The model developed by the JDL[23] (Joint Directors of Laboratories) has been selected for adaptation and refinement for the ATIS community for two reasons. First its functional representation is consistent with National and Regional ITS Architecture principles, guidance, and practices. Second, it is the most widely used model in the data fusion research and development communities. The JDL model is a multi-level, general framework allowing for attention and refinement of key system elements, such as objects, situations, and processes. Figure 4-1 depicts the ATIS data fusion model, with the following discussion highlighting general definitions and functions pertinent to ATIS.

Figure 4-1 A Data Fusion Model Applicable to ATIS

On the left, the model depicts input data received from multiple sources and/or sensors. Typical sources include roadway sensors (loops, cameras, infrared or microwave detectors, etc.), network intelligence sources such as probe vehicles (transit system AVLs or other private/public sources, as available), and traffic control system data and historic databases. Supportive ATIS information may include environmental conditions (weather, topological data, etc.), roadway information (as-built construction drawings, geometrics, etc.), historical data about the ATIS users’ preferences and system performance, other system notifications (utility company information about nearby roadway repairs via web-enabled intelligent agents), and condition reports or inferences obtained from travelers through telephone calls, PDAs, e-mail communications, etc. The scale or scope of this information may be drawn from local, regional, national, or worldwide sources, depending on the particular function and traveler information sought.

In general terms, the five levels in the model depict the major data fusion activities, which are:

Level 0 Source Pre-Processing: alignment and estimation of signal-observable or object observable states (changes in frequencies at a loop detector, pixel changes on a CCTV data set, etc.) at the signal-level or pixel-data characterization.

Level 1 Object Assessment: estimation and prediction of entity states (vehicle, building, pedestrian, temperature, wind speed, etc.) on the basis of observations and source data from Level 0 and supporting databases. Missing data are also addressed at this point, once proper editing and alignment have occurred.

Level 2: Situation Assessment: estimation and prediction of an ensemble state on the basis of inferred relationships among the objects defined in Level 1.

Level 3: Impact Assessment: estimation and prediction of effects based on situations of planned or predicted actions by others.

Level 4: Process Refinement: adaptive data acquisition, processing, and management consistent with the overall purpose of the data fusion system.

On the right, external interfaces allow for human-computer interface (HCI) and the dissemination of data fusion results. The dissemination of results may occur through public address systems, mass media, Internet-based broadcasts, Internet Service Providers (ISPs), in-vehicle communication subsystems, and/or various handheld (wireless) devices such as telephones, PDAs, etc. The HCI also provides a means of making queries of the data fusion model as well as monitoring/evaluating the system performance through off-line observations and analysis.

Depending on the system design and objectives, the ATIS data fusion processing at a level could encompass [24]:

§ Level 0 -- Processes input data from all appropriate sources, including real-time roadway sensor information, weather sensors, cellular telephone traffic, incident reports, etc. Level 0 activities include data formatting, referencing, and/or normalizing as well as managing the scheduling and process management functions to ensure all input data is available at the same time for the next level of processing. Most of the activities in this level are concerned with signal processing, transmission, data storage, and process management activities as defined by the overall system architecture and goals. The effectiveness of data sharing and data quality will depend, in part, on the use of standards and protocols at Level 0. For example, standard XML schemas (or DOM-based approaches) will enhance the ability of Level 0 processing to parse and distribute elementary source data.

§ Level 1 -- Processes the refinement of the Level 0 data into object identification (what it is) and state estimation (where it is and when). Some of the Level 1 processing involves information with some uncertainty and so processes and techniques must be employed to improve on the estimation process. At other times, the certainty of the detection, classification, and estimation of the signal comes with high confidence and so a minimal amount of processing is needed to achieve an optimal identification or estimation. Level 0 and Level 1 activities converge when the data object are characterized and identified as signals or features. Not all Level 1 activities are necessarily automated as in the case of operators making a final determination on certain pattern recognition processes.

§ Level 2 -- Processes merge the results of the Level 1 processing with information from other sources, including human-system interaction or databases. These sources may include weather reports, historical traffic patterns for key segments of roadways, GIS network data, and workzone locations. Level 2 fusion results in the estimation and prediction of the system state (levels of congestion, travel times on the defined network, micro-scale roadway weather status, and similar systems state descriptors) on the basis of inferred or absolute relationships. Level 2 processing may also identify the circumstances or situation causing the observed data and events, i.e., known workzone locations.

§ Level 3 -- Processes are a subset of Level 2 activities. Level 2 involves estimating and predicting all types of relational states, while Level 3 involves predicting specific relationships of interest between a specific object (vehicle or individual) and the environment. For example, Level 3 processing might assess multi-jurisdictional network traffic flow patterns and other data with respect to the likely occurrence of a quick-moving rainstorm through a particular area and the subsequent impacts on regional traffic flow. This assessment might then provide en-route guidance information to the ATIS user.

Level 4 -- Processes are for planning and control, but not estimation. As such, Level 4 techniques involve assigning resources to tasks (operating system or database management controls), evaluating the need for additional data resources (storing and processing capacity and speeds), or modifying some of the process or system parameters.

Interrelationships among the three primary levels of data fusion processing are illustrated in Figure 4-2. In this illustration, a subgroup of sensors might represent real-time roadway loop induction sensors, while the remaining sensors might represent weather data gathered from nearby environment sensing stations. The Level 0 data processing may occur i) simultaneously or at different times and ii) centrally or in separate distributed processes and paths. For example, ambient temperature detection and Level 0 processing (sensor signal conversion to a data object/packet) may occur in the field, depending on vendor equipment and system architecture. Polling of the environmental data may occur every 60 minutes, whereas the polling of the loop sensors may occur every 30 seconds. The sensed objects are combined in a Level 1 association process to confirm object identity, as needed, and provide state estimation of the object, e.g., forecasted temperature for that sensing location and surrounding area for the next several hours. Once the object identity and state estimations are complete, the objects are passed on for Level 2 and 3 processing.

Figure 4-2 Relationships Among Data Fusion Processing at Levels 1, 2 and 3

Figure 4-2 Relationships Among Data Fusion Processing at Levels 1, 2 and 3 [25]

The following subsections discuss the various techniques and suggested guidelines associated with the five functional levels in the ATIS data fusion model.

4.2.1 Level 0 (Source) Processing and Techniques

The purpose of Level 0 processing is to transform data received from multiple sensors into common spatial and temporal reference frames.

Raw data from mechanical, opto/electrical, or similar sensors usually require some signal processing and refinement before using. Most vendors provide firmware that allows for a variety of functions associated with the signal processing and data alignment. Major activities include the signal collection and digitization (as needed), local storage of the digitized data, processing for levels of detection (thresholding or gating), adjusting for false alarms, scaling or other adjustments based on established calibration processes. Once the basic signal has been normalized it is usually formatted for transmission as a data element or group of data elements (including the definition of metadata), interfaced with the communications subsystems, scheduled and/or released for transmission based on the communications protocols and polling requests. Specific data alignment functions could include coordinate transformations (e.g., topocentric non-inertial references to geocentric inertial coordinates), time transformations (e.g., mapping from reported observations to actual physical events), and unit conversions. Depending on the sensor and its application within an overall architecture, a self-assessment procedure may also be part of the Level 0 processing.

The sensor accuracy requirements or specifications are usually established by the overall ATIS service and system goals[x]. A wide range of factors can affect the measurement accuracy, including the design of the sensor, its placement in the operating environment (offsets, height, etc.), the conditions under which it operates (weather, lighting conditions, etc.), and the life-cycle maintenance and upkeep. Even if the raw data is properly processed and transmitted for subsequent use, the usefulness in applying the data is dependent on the overall system architecture and processing. For example, temperature readings from an environmental sensing station may be captured by an agency only for historical purposes and not provided except on an hourly basis. But if ambient temperature readings, coupled with the potential for roadway icing conditions in the vicinity of the sensing station, were to be properly fused, the resulting decision could affect traveler information, route selection, and safety.

Data quality is an important system performance characteristic at this level, since data fusion methods will be only as good as the raw data that is supplied. The data sensor quality is affected by a number of parameters, including sensor specification, sensor locations, operating environment, polling frequency, and signal processing. To enhance data quality, some sensor subsystems are including metadata elements to aid in qualifying the raw data quality. For example, in certain inductive loop detectors, a set of metadata descriptors are provided by the sensor that indicate not only the data format of the transmitted data, but error checking information (indicating if the sensor working, if it has been calibrated, etc.). This information can be further processed during Level 1 and Level 4 activities to improve estimations. ATIS data definitions and NTCIP standards provide technical guidance on Level 0 data quality, primarily from the transmission and communication perspective.

The strategy for handling missing data should be established during the design and calibration of the data fusion model since these are the points at which an overview of all available data sources can be made to determine the desired levels of data precision, reliability, and the best data adjustment methods to minimize distortion and maximize the usefulness of the substituted data. The approach for handling missing data should satisfy four criteria: i) limit the biases caused by not having a complete and accurate record; ii) contain an audit trail for evaluation purposes in Level 4 activities; iii) ensure the substituted/missing data values are internally consistent with the overall design and intent of the data fusion system; and iv) be objective and efficient. A range of missing data techniques is available, such as substituting the last known value (from a time series, for example), estimating the missing value with a parametric model defined during the data fusion system design and calibration phase, making an inquiry for the missing values through the Human-Computer Interface, tagging substituted data values, or alerting subsequent data processing functions to avoid calculations involving this (non-critical) data element until the missing data is resolved. These represent five common methods, but whichever method is selected, it must be internally consistent, efficient, traceable, and objective.

Not all data may come from field sensors, as described above. For example, an ATIS data fusion model may seek information from other websites or ATIS users about network conditions or other data needed to enhance the ATIS needs. In this case, data alignment and protocols/standards become critical requirements.

4.2.2 Level 1 (Object Assessment) Processing and Techniques

Level 1 processing is conducted to achieve two purposes: object identification and state estimation of the object or its “track” (time and changes in location or representation) The identification and state estimation may occur simultaneously, however the following discussion keeps the discussion separate for clarity. The outcome from Level 1 processing is a “situation statement” for each individual object, e.g., a vehicle object captured by a CCTV image, a digital phone record of a cellular caller’s report of an incident, or a temperature reading at a specific milepost location.

Sensor outputs from Level 0 are available to various algorithms to help estimate the object identify and its track. Level 0 signal data may indicate vehicle occupancy over a loop sensor while Level 1 processing helps confirm the presence of a vehicle and possibly its classification, depending on the type of sensor. The degree of accuracy and performance of Level 1 processing is established as part of the overall ATIS system design and usually includes such design elements as the required sensor quality (expressed within certain confidence levels, such 95% or 99%), availability (overall system up-time is 95% or greater), and timeliness (key object data is reported within seconds or minutes, as appropriate).

Implications for data quality at Level 1 can be delineated in terms of a variety of metrics, such as positional accuracy, completeness, validity, consistency, and timeliness. Positional sensor accuracy is based on subsystem functioning of the sensor, the communication system, and the processing of the sensor data, a sequential process that may cause a cumulative degradation in data quality if any sub-element is defective or performing poorly. So while a sensor may be functioning at a 99% confidence interval, the processed data may only be 50% reliable due to communication system faults, for example. Data completeness refers to the percentage of Level 1 data fields that have values entered, namely for which the data collection (Level 0) and assessment processes (Level 1) successfully occurred. Data quality for completeness is usually a “yes” or “no” indicator with a threshold established for the required percent of completed data field[y]. Validity refers to the percent of data having values that fall within respective domains or allowable values. As with data completeness, data validity is assessed by means of an acceptable threshold of allowable values. Consistency refers to the stability of the Level 1 assessment process and is usually evaluated using the percent of matching values (within tolerances) across data records. Timeliness refers to the extent at which the data item is provided at the time required or specified by the data fusion system. Level 4 functions of the data fusion model is able to monitor and assess the timeliness of Levels 0 and 1 data quality through exception reporting, data or process tagging, and other process monitoring techniques.

4.2.2.1 Object Identification Methods at Level 1

For Level 1 identification purposes, objects are usually classified hierarchically into one of four categories, as shown below in Figure 4-2[z]. The purpose of the fusion processing at Level 1 – object identification – is to estimate the object category based on the sensed data.

The level and accuracy of the discrimination among the categories depends on a number of factors including the type of sensor, the resolution of the sensor, any a prior knowledge about this sensor (such as bias, Mean Time Between Failure), and the signal-to-noise ratio at the input to the sensor. Also, the object may be well defined and identified if it comes from a “trusted” data source, e.g., a set of roadway design specifications from an agency’s database which have been formatted, verified, and quality controlled. In this case, detailed object identification activities are not required and the data object is forwarded in support of other data fusion algorithms and activities.

Figure 4-2 Levels Of Object Identification

The major mathematical techniques to achieve object identification can be grouped into three broad groups with corresponding subgroups: physical models, feature-based models, and cognitive-based models.[26],[27] Figure 4-3 illustrates the range of major techniques, with the subsequent discussion starting on the more conventional and established state-of-the practice methods, followed by newer, but feasible techniques.

Physical Models	Feature-Based Models		Cognitive-Based Models
§ Kalman Filtering	§ Parametric Techniques	§	§ Logical Templates
§ Maximum Likelihood estimators	§ Non-Parametric techniques	§	§ Knowledge-Based Expert Systems
§ Least Square Approximations		§	§ Fuzzy Set Techniques

Figure 4-3 The Major Techniques Appropriate For ATIS Object Identification (Level 1)

4.2.2.2 Physical Models For Object Identification[aa]

Physical models take the sensor-observed data and estimate the identity based on matching algorithms that compare modeled or pre-determined/pre-stored object descriptors with the observed data. Examples could include inductive loop sensor data as a function of vehicle types, temperature, or height profile images. The estimation techniques for physical models are primarily simulation and estimation methods. Estimation processes, such as Kalman filtering, maximum likelihood estimation, and least squares approximation, are representative methods and can be considered state-of-the practice.

4.2.2.3 Feature-Based Methods For Object Identification

Feature-based inference techniques do not use physical models. Instead, correlation is performed by mapping the observed data into an identity declaration. Feature-based algorithms can be subdivided into parametric and non-parametric or what Waltz and Llinas³ refer to as information theoretic techniques.

Parametric techniques. The identification challenge is addressed by mapping (through estimation techniques) the parametric data directly into a declaration of identity. Parametric techniques require an explicit “measure of match”, such as a least squares estimation. Parametric techniques include classical inference, Bayesian inference, Dempster-Shafer inference, and Generalized Evidence Processing.

Classical inference starts with an assumed hypothesis and then seeks to determine the probability that an observation can be identified with or attributed to an object or event. Metrics associated with these techniques include determining confidence intervals, calculating sample size for a desired margin of error, conducting test for statistical significance, etc. Decision theory can be used to select among hypotheses. The major disadvantage with classical inferences for ATIS services is the collection of representative probability density function to assist in classifying the observation into a category. As multivariate calculations are proposed, the complexities in defining joint probability density functions increase substantially.

Bayesian inference addresses some of the challenges of classical inferences. Classical inference allows for the calculation of an event probability based on a hypothesis. Bayesian inference provides a means of determining the probability of a hypothesis being true given the observed data. For example, Bayesian inference could answer an important question such as ‘what is the likelihood a cellular caller can accurately verify the location of an incident given the history of incidents in that area?’. Multiple hypotheses can also be assessed.⁵ The drawbacks of the Bayesian inference technique include the challenge of defining prior likelihood functions and the complexities of disaggregating and estimating when there are multiple hypotheses and conditionally-dependent multiple events.

The Dempster-Shafer (D-S) method is a generalized version of the Bayesian inference method for assessing propositions, i.e., that an object is in a category. Through its generalization, D-S can account for general uncertainty by distributing the required evidence (known as support) for a proposition between the proposition and the union of any propositions that include it. Any evidence that cannot be directly assigned to a proposition (or its opposite) is assigned to an uncertainty set. The support set and the uncertainty set make up the hypothesis space. Bayesian and D-S yield the same results when all the basic propositions are mutually exclusive and no support is assigned to the uncertainty set. The challenges of the D-S method include the formulation of the hypothesis, which can become complex with only a few variables, and the assignment of the degree of support for a proposition. Klein provides a more detailed example of the D-S application for ATIS[28].

Generalized Evidence Processing (GEP) uses evidence collected by multiple sensors and assigns a probability mass to the evidence, much like the D-S technique. GEP differs from D-S in that the probability mass assignments and their combination are based on the a priori conditional probability of the propositions or hypotheses⁵. GEP has a similar set of implementation challenges as D-S.

Information theoretic (non-parametric) techniques. The parametric techniques listed above incorporate an explicit stochastic aspect of the observed data into the calculations. Information theoretic techniques make no attempt to model stochastic features of the observed data during the process of object identification. The information theoretic techniques include parametric templates, artificial neural networks, cluster algorithms, voting methods, and Figures of Merit, pattern recognition, and correlation measures.

Parametric templates are one of the most straightforward means of object identification. Sensor data, collected over a pre-determined period of time, are combined with other information sources to determine if they match pre-selected conditions and thereby identify an object. Parametric templates can be used for object/event detection, general situation assessment, or single object identification.

Artificial neural networks are combined hardware or software subsystems that emulate the thinking processes that occur in biological nervous systems. The technique involves defining the various system states, establishing processes that map input data or stimuli into intermediate or output categories/identity states, and conducting sufficient training and initialization to allow the network to become sufficiently accurate and self-sustaining.

Cluster algorithms take data elements and groups them into natural sets based on an association metric. The association metric is used to describe the closeness or nearness of fit among the data and the clusters. Association metrics may be spatially defined or statistically defined. The meanings of the clusters are usually the result of an expert or team providing an interpretation. In general, clustering algorithms and the interpretation of the clusters are subject to variability in the final object identification due to data preparation (scaling, data cleansing, etc.), the choice of association metric and algorithm, and possibly the order or sequence of the input data or sets. Hence, application of cluster methods must be judged on their overall effectiveness, post-assessment performance, and reliability to form consistent and meaningful object identity clusters⁶.

Voting methods treat each sensor's data declaration as a vote in which majority, heuristic rules, plurality, weighted voting, or other forms of decision-tree rules may be applied to determine the final object identity. The challenges with voting methods are the training time and the declaration of the detection modes and desired confidence levels. Calibration is an iterative process in which the relation of detection and false alarm probabilities are assessed against desired identification confidence levels.

Figures of Merit (FOM). Observations are compared with each identity category and with other observations. Because the observations can contain both parametric data (time, location, etc.) as well as non-parametric data (e.g., an identify declaration from a sensor), then a figure of merit is used to qualify the degree of association with the category. The FOM is a numerical measure of the confidence the two observations represents the same object/entity. Numerical measures are similar to the distance measures used in clustering algorithms, such as Euclidean or Mahalanobis. When working with alphanumerical data, a different approach is used, usually based on heuristic insights rather than more rigorous mathematical measures.

Correlation measures are derived from the combinations of weighted figures of merit. Correlation measures allow for a comparison score or measure of correlation to be assessed across systems that have numerous figures of merit. Used in this fashion, the correlation measure represents the total likelihood that two objects/entities are the same.

Pattern recognition is used for the description or classification of sensed data[29]. Three primary approaches to pattern recognition are currently in use. First is statistical pattern recognition (sometimes referred to as a decision theoretic approach) in which a set of characteristic measurements or features is extracted from the input data and use to assign the set of features to one of several pre-determined classes. This technique assumes a class is pre-defined by an underlying statistical model that represents a state of nature, set of probabilities, or probability density functions to which the features can be matched or assigned. Second is syntactic pattern recognition (also referred to as structural modeling) in which the technique focuses on the relationships or interconnections among features in order to yield identity information. The structural similarity of patterns is assessed by quantifying and extracting structural information using, for example, the syntax of a formally defined language, e.g., digraph theory. Third is the artificial neural network approach applied to pattern recognition, which was discussed previously.

4.2.2.4 Cognitive-Based Methods For Object Identification

Cognitive-based models attempt to mimic the inference processes of human analysts in recognizing object identity. The most widely used techniques include logical templates, knowledge-based expert systems, and fuzzy set theory.

§ Logical templates use a comparison process in which a predetermined and stored pattern is matched against observed data to infer the identity of the object. Logical templates are also useful in assessing the Level 1 position estimation (another key function in the Level 1 processing) as well as constructing an overall situation report (Level 2 processing). Templates can be developed for both parametric and non-parametric data. Identity can be established through Boolean relations as well as through relative measures of association. Semantic logic, which is an evolving field, may also be employed to assist with classification. Fuzzy logic can be applied to the pattern matching technique to account for a range of uncertainty in either the observed data or the logical relationships used to define a pattern.

Knowledge-based expert systems incorporate formal rules and other knowledge from experts, through a process of knowledge engineering, to automate the object identification process. Knowledge-based expert systems usually consist of five subsystems: a knowledge base that contains facts gleaned from one or more experts through carefully structured interviews and data collection processes; description of specific algorithms and a set of heuristic rules that reflect the processing steps of the expert(s) for the knowledge domain; a database; a control structure or inference engine; and a human-computer interface. The inference engine processes the data by searching the knowledge base, applying the facts, algorithms, and rules to the data, and suggesting a set of actions, usually presented at the human-computer interface. Completely automated knowledge-based systems involve decision-making among the set of actions without the use of the human-computer interface. Intelligent agents and genetic algorithms are typically considered to be subclasses of knowledge-based systems.

Fuzzy set theory and techniques. Fuzzy logic consists of a variety of concepts and techniques for representing and inferring knowledge that is imprecise, uncertain, and unreliable. Fuzzy logic can create rules that use approximate or subjective values and incomplete or ambiguous data or relationships. Application of this technique (at Level 1) will allow for a mapping of data or observations into appropriate identity categories. To apply fuzzy logic requires three subsystems: a definition of the fuzzy sets, the membership function (fuzzification), and a set of production rules that are represented in the form of If-Then rules. The logical processing using fuzzy set is known as fuzzy logic. Fuzzy sets incorporate vagueness into the production rules, since they can represent less precise linguistic terms (e.g., hot/cold, short/long, etc.). The production rules operate simultaneously and influence the output of association and category definition.

These techniques represent the more practical set of methods and techniques for Level 1 object identification. There are additional fields of study that are emerging, but these are still in the development stages and include multiple criteria optimization, multiple hypothesis testing using knowledge-based guidance, random set theory, conditional algebra, and relational event algebra.

4.2.2.5 State Estimation Methods at Level 1

State estimation seeks to combine parametric data from multiple sensors to obtain the most accurate estimate of an object’s state and change of state. The state estimation is achieved through a combination of physical models (such as equations of motions or observational models) and statistical assumptions about the observation process to match the observed data to a state vector, a set of variables such as position and velocity that can be used to predict future states (“tracks”). The tracks are stored in a central track file for use in estimation of subsequent state values. For ATIS, the primary areas of interest for state estimation include vehicle tracking (velocity and position), situations affecting traffic flow (incidents, workzones, ramp metering performance, toll plaza/booth performance, etc.), overall roadway conditions (mainline and arterial levels of performance), and weather advisories. State estimation algorithms can be used at the level of the sensors and/or in software utilized at traffic management centers or other data collection and processing locations to make estimations.

Two broad categories of state estimation and tracking have been identified^5,6. The first category is based on a search direction identified by the sensor (data) or object (target). The second category is based on techniques needed to associate and correlate data and state vectors.

Direction tracking systems can be either sensor driven or object driven. In sensor-driven systems, state estimation data associated with an observed object (e.g., a vehicle) are used to initiate a search through a track database for other tracks that can be associated with the state estimation data according to some level of confidence. Object-driven systems works somewhat in reverse by first identifying a primary tracking sensor (based on the track) and then managing other sensors to acquire data or search databases that can be associated with the track. Sensor-driven approaches are the most common ATIS data fusion method employed, since active sensor control may be difficult to achieve across institutional and operational jurisdictions.

§ Association and Correlation of Measurement Data and Tracks. The association and correlation of measurement data (e.g., position, velocity, temperature, etc.) and tracks from multisensor inputs ultimately create central track files with some prescribed level of confidence. In an ATIS data fusion system, it is desirable for each track file to represent a unique physical object, which has significant implications for database management.

To construct this set of state estimation track files, a number of specific processing steps are required: data alignment; prediction or threshold gates; association metrics; data and track association; position and kinematic estimation; and attribute estimation.

Additional data alignment may be required at Level 1 to make transformations through spatial and temporal reference adjustments and coordinate system to ensure a common space-time reference for state estimation. Errors introduced by measurement inaccuracies, adjustments for levels of precision during coordinate transformations, and object “noise” can be accounted for (to some degree) during data alignment.

Prediction or threshold gates are used to eliminate or gate certain types of associations based on known properties or estimates of the object. For example, rapid acceleration of vehicles during a sampling and observation period may not be possible and gating would eliminate such associative considerations. The magnitude of the gate reflects the calculated or expected object state errors associated with the algorithms used to make the calculation.

Data association metrics quantify the similarity of the observations. A variety of standard mathematical measures are available, including Euclidian and Mahalanobis, or variations of those.

Data association occurs by applying appropriate metrics to compare tracks and measurement data. The data and track association technique is dictated by the type and associated complexity of the tracking problem (such as vehicular speed and temperature changes) as categorized by single object–single sensor, single object–multiple sensor, and so on. Some of the association techniques discussed in the object estimation discussion are applicable to state estimation, e.g., maximum likelihood estimation and Bayesian inference.

4.2.3 Summary of Level 1 Methods

The ATIS data fusion model is a non-trivial design challenge involving not only the selection of appropriate methods and algorithms, but careful attention and focus on the identification of the problem to be resolved through data fusion. The specification of the data requirements is but one component of this complex design process, which is discussed further in Section 5.

The techniques listed for Level 1 data fusion require a variety of data and information, all bound by the specific mission for which the data fusion model is being developed. Data fusion techniques, such as Bayesian inference or Dempster-Shafer require expert knowledge or information from the data fusion system designer to define (or have experts define) the appropriate a priori probabilities and likelihood functions, the probability masses, and desired confidence levels. Similarly, the use of fuzzy logic applications will require the designer to develop the appropriate membership functions and production rules based on a knowledge engineering and information extraction process. Other algorithms, such as classical inference, knowledge-based rule systems, and pattern recognition, also require the designer to assume probability density functions, rules, or other parameters for their operation. Implementation of these data fusion algorithms is thus dependent on the expertise and knowledge of the data fusion system designer, an understanding of the data fusion mission, a proper analysis of the overall operational situation, and the types of information provided by the sensor and other sources.

Figure 4-4 illustrates the relative qualitative merits of the Level 1 techniques.

	Relative Performance	Scalable	Computational Complexity (Time)	Maintenance	Cost to Implement
Parametric Based
Classical Inference	Excellent	Excellent	Excellent	Excellent	Excellent
Bayesian Inference	Good	Poor	Good	Poor	Poor
Dempster-Shafer	Good	Poor	Good	Poor	Poor
GEP	Poor	Poor	Poor	Poor	Poor

Non-Parametric Based
Parametric Templates	Poor	Good	Good	Poor	Poor
Neural Nets	Good	Good	Poor	Poor	Poor
Clustering	Good	Excellent	Good	Good	Good
Voting	Good	Excellent	Excellent	Good	Excellent
Figure of Merit	Good	Good	Good	Good	Good
Correlation Measures	Excellent	Excellent	Good	Good	Excellent
Pattern Recognition	Good	Poor	Poor	Poor	Poor

Cognitive Based
Logical Templates	Poor	Good	Poor	Poor	Good
Knowledge-Based	Poor	Poor	Poor	Good	Poor
Fuzzy Set Techniques	Good	Good	Good	Good	Good

Figure 4-4 The Relative Merits of Level 1 Data Fusion Techniques

4.2.4 Level 2 (Situation Assessment) Processing and Techniques

The creation of the situation assessment (Level 2) is an iterative process of fusing spatial and temporal relationship among entities to group them together to form an abstracted and probable interpretation of objects associated with the travel context. Development of the situation assessment requires the production and maintenance of an appropriate, multi-level abstraction of a dynamic situation. At this point, the data fusion process could be compared to assembling a complex jigsaw puzzle for which no clear picture or only a partial picture of the completed scene exists. Level 1 analysis offers insights into vehicle identification, vehicle movements, special events or activities.

Key functions and techniques of Level 2 processing include the following2:

Object aggregation involves the establishment of relationships among objects including temporal, location, and entity dependencies (functional, associative, precedent, mathematical). For example, vehicular density, average speeds, and vehicle types for a roadway segment could be the result of object aggregation. Techniques for this function include hierarchical classification methods, neural nets, nodal graphs, and spatial relationships.

Event and activity aggregation involves the use of a priori reasoning to establish temporal relationships among entities in order to identify meaningful events or activities. For example, a collection of Level 1 environmental sensing station objects (surface and air temperatures, wind speed, wind direction, humidity) and state estimations (roadway surface temperatures and local wind patterns) can be used to define roadway segment environmental conditions.

Techniques appropriate to event and activity aggregation come from the knowledge-reasoning field. Three classes of techniques are available: problem-solving paradigms (e.g., rules, procedures, genetic and neural algorithms, and statistically based algorithms), evidence combination strategies (e.g., Bayesian inferences, Dempster-Shafer, and fuzzy set theory), and decision-making approaches (e.g., rule instantiation and parametric algorithms).

Contextual knowledge and interpretation involves analyzing objects and state estimations with respect to the context of the current situation assessment. For example, the presence of deciduous trees (image data) in the summer (based on date information) would allow for the deduction that the possibility of a deep snowfall in an associated roadway segment is highly unlikely. Contextual knowledge may be provided exogenously to the fusion model or be derived from other contextual non-sensor-driven data contained in supporting databases, such as terrain, historic traffic flow patterns (including seasonal and day-of-week variations), workzone locations, special events, and emergency situations.

4.2.5 Level 3 (Impact Assessment) Processing and Techniques

Level 3 assessment focuses on the possibility and probably outcomes associated with a specific event or action. As mentioned earlier, Level 3 can be considered a subset of Level 2 activities and functions. Level 3 functions are usually implemented as a specific prediction in a topical area (such as workzone traffic flow, intermodal operability/performance, network congestion) , drawing focused types of inferences from Level 2 associations. The impact assessment outcomes may be characterized by such parameters as the impact likelihood or cost/utility measures associated with the potential outcomes of an inferred action or probable event. Techniques for Level 3 rely primarily on knowledge-based reasoning systems, mentioned in previous discussions.

4.2.6 Level 4 (Process Refinement) Processing and Techniques

Level 4 processing involves planning and control functions, not estimation. Activities include the monitoring and evaluating of the data fusion model and corrective processes to refine the algorithmic and estimation process, managing databases to ensure optimal system performance, and regulating the data acquisition to achieve optimum results, which may involve direct sensor control, data caching/batching, and selecting/de-selecting data sources. Specific functions include:

Identifying changes or adjustments to processing functions within the data fusion domain that may result in improved performance (self-assessed or based on mission benchmarks)
Determining and seeking specific data requirements (from specific sensors, reference data, etc.) needed to improve individual and combined level outcomes or estimates.
Recommending allocation and direction of resources (sensors, CCTVs, variable message signs, etc.) to achieve overall mission and goals of the data fusion system
Employing advanced database management techniques to provide monitoring, evaluation, updating, plus data processing functions such as merging, purging, retrieval, searches, substitutes, etc.
Managing global processes, such as time and spatial synchronization across communication systems, sensor collection points, metadata compliance, and databases.

ATIS data fusion Level 4 activities currently only include the most basic database monitoring and communication check functions. Advanced sensors or smart sensors, allowing for centralized control of field equipment, have only been tested experimentally.

4.3 Data Fusion Architectures

Architecture refers to a structure of components, their relationships, and the principles and guidelines governing their design, implementation, and evolution over time. Data fusion architecture involves four fundamental components and their interrelationships: the data sources, the data fusion algorithms and database techniques, the communication networks, and the HCI interface.

As expected, the specific configuration of the data fusion architecture is a complex design process involving tradeoffs among all of these components in a cost effective manner that meets the ATIS system goals for function and performance. Because of the wide variety of ATIS fusion applications and applicable architectural components, it is not possible to provide a prescriptive, detailed definition of which architectural components and fusion techniques are best. Consequently, this discussion provides suggested guidelines for the four fundamental components of the data fusion architecture.

4.3.1 Architecture Implications -- Data Sources

Data sources may come in a variety of formats through various communication channels. Three basic architectural alternatives can be used to capture multi-sensor data sources: direct fusion processing of sensor data; representation of sensor data using feature information, with subsequent centralized processing of feature vectors/arrays; and processing of each sensor to achieve a high-level of inference/decision about an object4.

The first configuration involves a direct fusion of data at the sensor level. This would occur when the sensor is able to perform a substantial number of the basic fusion functions, e.g., data alignment, object association, and others. Moreover, if the same set of sensors are measuring the same physical phenomena, then the sensor data can usually be combined directly. This might be the case with a series of loop detectors for a roadway segment[bb]. Techniques for raw data fusion typically involve classic estimation methods such as least-squares and Kalman filtering. If the sensor data are not the same or incommensurate, then the data should be fused using feature information or at the inference-decision level.

The second configuration uses feature-level fusion. Features are an extraction of a representative feature from the sensor data, such as a set of regression coefficients or Fourier transform coefficients. Features are extracted from the multi-sensor observations and combined into a representative feature vector that is typically offered to a pattern recognition technique, such as clustering algorithms or template methods. Examples of this approach include the used of CCTV and acoustic sensors to identify and estimate vehicular flow data in a critical highway segment.

The third configuration combines sensor information after each sensor has made a preliminary determination of an object’s identification and location. Techniques for decision-level fusion include voting techniques and parametric inference methods. More advanced CCTV devices, especially color cameras, are offering object and location referencing based on this type of architecture. Similarly, GPS-enabled cellular phones use this approach to provide precise location information (within 3-10 feet) based on the caller’s device.

There is no clear-cut preference for a data source architectural alternative. In fact, most data fusion architectures will be a hybrid of one or more of these configurations. For example, a combination of CCTV and supporting image detection sensors may best be configured for a feature-level design intended to support parking management. Specific design features would need to address such issues as cameras resolution, refresh rates, and aperture field.

4.3.2 Architecture Implications -- Data Fusion Algorithms and Database Techniques

The relative merits of various data fusion techniques for the multi-level data fusion model were discussed in previous portions of this section and summarized in Figure 4-4. When viewed from an architectural perspective, the numerical and heuristic techniques used to perform data fusion will vary widely depending on the environment in which they are embedded along with the available functions and source data. As noted, the algorithms involve substantial computations to make association, correlations, estimations, and classifications of objects. To make these calculations and estimations, the data fusion algorithms perform operations on or with model parameters, sensed data, external database information, performance data, and a priori data according to a specific technique such as Bayesian inference or knowledge-based systems. These operations almost always require the use of databases for data input, storage, retrieval, archiving, and other functions. Consequently, database management is a key consideration in the overall data fusion system architecture and performance.

Most databases available for data fusion are either relational databases or object-oriented database (OOD). Relational databases are relatively efficient in well-defined basic functions such as estimation, sorting, storing interim calculations, or assembling track files. Hence data-centric ATIS services are more likely to be implemented using relational database architectures if judged primarily by cost effectiveness and processing performance criteria. More complicated fusion functions, such as association and classification, may require computations better suited to object-based data structures since the data fusion algorithms will need to explore complex relationships among many types of data objects. Relational databases are not as effective in multi-dimensional analysis as OOD. Model-centric ATIS services would likely be implemented on an OOD configuration if processing flexibility and sophisticated fusion objectives were the primary criteria. Hybrid database products that combine the best features of relational databases and OOD concepts are also emerging, but are usually selected for narrow, application-specific needs and supported by highly skilled staff.

Another database design consideration affecting data fusion system performance is the extent of centralized versus distributed databases. Almost all ATIS data fusion architectures will be migrating into distributed database configurations for three reasons. First, a single agency or organization does not possess all of the data required to support ATIS users’ needs. Second, distributed databases allow for distributed computing, more effective resource sharing, cost effective specialization, and in some cases, system redundancy. Third, the communication network connectivity and performance (speed, accuracy, security) and associated costs are favorable in relationship to the benefits derived from a distributed architecture.

Solutions to address distributed computing have spawned the basis for client/server and n-tiered architectures in which middleware provided connectivity and interface management. Enterprise-focused database practices have resulted in the development of the data warehouse concept in which all corporate/agency data resides and is accessed by a number of software applications. Data marts are subsets of the larger data warehouse and are usually created to summarize or extract information needed for a specific purpose. Data fusion operations may involve parallel interactions with multiple data marts to allow for more rapid access and processing speeds associated with fusion-specific algorithms.

As the distributed architectures and client-server configurations increased, especially across the world wide web-based networking infrastructure based on TCP/IP, data management problems have increased. For example, distributed database management for data fusion may require connectivity among distributed homogenous databases as well as distributed heterogeneous databases. Consequently, middleware techniques have expanded to include object-oriented semantics to address the interfacing issues, such as interface standards (and backward compatibility), access privileges, metadata uses, and data quality. A wide range of techniques has emerged to address the interface issues, sometimes referred to as tightly coupled or loosely coupled. Industry groups and standards organizations have worked to develop comprehensive solutions (tightly coupled) in which interface component are well defined, hierarchically organized, and application software developed according to these specifications. Other groups have only defined the desirable structure and requirements for selected interface components[cc]. While there are a variety of methods for managing distributed databases[dd], two common means of handling distributed database computing for ATIS are Common Object Request Broker Architecture (CORBA) and a Distributed Common Object Model (DCOM). Both CORBA and DCOM achieve language independence by defining object interfaces in a language-independent manner[ee]. Hence data fusion algorithms can use distributed objects transparently, although with higher processing overhead due to the translation and exchange requirements. Single-vendor servers, integrated with back-end databases, have helped to improve the efficiency and timeliness of OOD, however these improvements come at the expense of multi-vendor interoperability and interchangeability. The key design tradeoff for data fusion database configurations focus on the relative processing speed of the object interface and the alternative communication network protocols for exchanging the data packets.

CORBA and DCOM are examples of object-oriented techniques applied primarily to the field of distributed database management. In an analogous fashion, Document Object Model (DOM) is applied to the area of dynamically accessing and updating the content, structure, and style of web-based documents rather than databases. DOM is being developed by the W3C group with the desire to have a platform and language-neutral interface for document access and updates. Employing object-oriented techniques, DOM utilizes a multi-level model for i) navigating and manipulating documents, ii) defining style sheet models and methods to manipulate the style sheet, and iii) document loading, updating, and saving. Commercial software is emerging to support DOM applications[ff].

4.3.3 Architecture Implications -- Communication Networks

As mentioned earlier, the NTCIP is the preferred communication architecture for ITS applications. Three types of communication are required for data fusion: file transfers, C2F, and C2C. ATIS data fusion architecture is most affected at the NTCIP information and application layers because these represent the newer components in the overall communication architecture. The lower layers (transport, sub-network, and plant) employ widely accepted and field tested industry standards, exhibiting well-known performance and insights into network design tradeoffs. The ultimate performance of the full communication network will depend on a complete definition and assembly of all components, allowing for an assessment of the system performance using such key criteria as speed, accuracy, security, availability, flexibility/scalability, standards conformance, standards maintenance, and cost to implement and maintain. Simulation tools such as OPNET are typically employed to evaluate the various communication network design options and performance tradeoffs.

The first set of NTCIP applications layer protocols pertain to file transfers, which are handled through well-established and readily-available protocols such as File Transfer Protocol (FTP) or Trivial File Transfer Protocols (TFTP). The design choice is determined by the desire for a connection or connectionless type of transport layer. Connection-oriented communication provides greater guarantees and transmission checking than connectionless-oriented configurations. The choice is dependent on the criticality of the ATIS information being exchanged, speed, and cost.

The second set of NTCIP application layer protocols pertain to C2F communications, which typically involve communication between a traffic or transit management center and field equipment. NTCIP allows for relatively straightforward communication protocols for exchanging data objects. These may be done with well-established protocols such as SNMP, which is suitable for traffic demands with high bandwidth and low volumes of messages, e.g., certain types of high-resolution CCTVs images with low polling frequency. Cellular phone systems and certain OOD systems employ SNMP as a means of managing data packet transmission and monitoring/managing selected network functions. STMP, developed by the NTCIP, reduces packet overhead with more efficient data encoding rules[gg] and is most suitable for low bandwidth and high volumes of messages, e.g., traffic signal systems. STMP is restricted however to 13 message sets, but more are under development.

The third set of NTCIP application layer protocols pertain to C2C data exchange, which is managed through DATEX-ASN.1 or CORBA. DATEX-ASN.1 employs a relatively simple procedure for exchanging data and is a cost effective solution for low bandwidth, small system needs. DATEX-ASN.1 is not however an object-oriented protocol. CORBA is able to exchange data, including objects, as well as activate methods embedded in remote objects, i.e., initiate remote processing. CORBA provides a full range of features for data exchange (both data and objects) but requires increased message overhead, resulting in higher implementation resources, including technical skill levels and maintenance costs. Consequently, CORBA is selected when a management center needs high communications bandwidth, the message volumes are relatively high, and the data and data processing methods are essential or mission-critical to the center’s operation.

4.3.4 Architecture Implications -- Human-Computer Interface

The HCI is an important data fusion architectural component. The most obvious HCI design considerations are with the ATIS user. The technologies and human factors features of HCI devices for ATIS applications have come from insights gained in other fields, namely industries such as PDA retailers, the wireline and wireless telephone industries, computer retailers, vehicle manufacturers promoting telematics, pattern recognition retailers, and the commercial mass media, which helps to shape content and distribution methods. ATIS user preferences and their implications for HCI interface design are still an evolving area. For example, the level of consumer acceptance of interactive voice recognition (IVR) techniques, used in the travel, banking, and the telecommunications service industries, is still an ongoing debate. Recent cellular phone products and 511 deployments have started to promote IVR as an improved, safer method of operating devices and communicating.

Other architectural aspects of the HCI involve the actual human interface with function points in the data fusion model. These involve standard computer-based interfaces which allow for system control and data update. Certain data fusion algorithms are not completely autonomous and therefore may require human judgments at certain points in their calculations and analysis. Consequently, the HCI must provide alerts and opportunities for operators to supply answers to interim estimations or inferences, propose hypothesis or tests through, for example, SQL inputs, and make annotations on the data fusion system performance.

4.4 Data Quality Management And Assessment

Data quality for ATIS data fusion is a multidimensional issue. Just as a vehicle has many quality dimensions associated with it, a data element or information product also has quality dimensions. This concept is important in developing methods to assess and improve ATIS data quality, since different levels of importance and different organizations/owners place potentially wide-ranging values on the dimensions of data quality. For example, some organizations or data contributors may be data-centric while others are model-centric. A public agency may be very satisfied with their level of data quality to estimate mean speeds. A third-party ISP may need greater resolution in sensor data in order to meet their business objectives, but may not be able to achieve it from a public agency. These differences may lead to a partnership for enhancing input data quality. However, to be successful in achieving effective data quality management, attention to all the quality dimensions is required by all data owners and users.

In general, data quality can be defined as fitness for use by information consumers[hh]. The degree of fitness pertains to five main quality dimensions illustrated below in Figure 4-5. The techniques or perspectives listed on the right-hand side of the figure are defined as follows:

§ Systems Planning view – Separate data classifications or classes from one another in order to support strategic, tactical, and operational assessments. Strategic and tactical views focus on the role that data has in the overall system purpose and types of decisions or information to be produced. Operational views examine the transactional data needed to carry out the routine activities of the fusion model.

§ Function view – This is the most common type of perspective and represents the functions to be performed with the data, e.g., association, correlation, estimation, and classification.

§ Applications System view –A focus on the entity-relationships among applications and interfaces within the context of an enterprise data model.

§ Process view – A perspective that indicates how the data is consumed or transformed throughout a multi-stage, multi-level process. The process model is typically used to validate the objects and object transformations in a data model, and thus assess data quality.

§ Time-Dependent view – A high-level perspective of the current (As Is) and future (To Be) state of data quality, with defined improvements in data definition, architecture, applications, business processes, policies, and objectives to achieve improved data quality as characterized by the future state.

Data Quality Category	Data Quality Dimensions	Techniques or Perspectives For Assessing
Intrinsic value/quality	Accuracy, objectivity, believability, reputation	Systems Planning view Functional view
Contextual or domain quality	Relevancy, value-added, timeliness, completeness, amount of information	Application systems view Process view Time-Dependent view
Representation or presentation quality	Interpretability, ease of understanding, concise representation, consistency, ease of manipulation	Functional view
Data Models	Accuracy, granularity, comprehensiveness, robustness, consistency	Functional view Applications systems view
Information Policy	Accessibility, access security, ownership, metadata, redundancy, cost	Systems Planning view

Figure 4-5 Data Quality Categories, Dimensions, And Techniques for Assessing

Many data quality improvement techniques or programs focus on the first or second dimension, the intrinsic value/quality or domain value. Typically techniques focus on the functional view or process view to find and fix the data errors. Findings errors associated with intrinsic or domain values can be done a number of ways, but almost all generally involve identification and correction. Identification is done by locating missing data, inconsistent data or outliers, duplicates, or traceable error due to erroneous conclusions or reasoning, e.g., Type 1 or Type 2 errors. Correction of erroneous data values may be done through many techniques, but they tend to employ domain-dependent methods that use keyword or domain relevance substitution, merge/purge, or data combination/reduction[30]. These “find and fix” techniques are used when there is a high understanding of the context in which the data is to be used, the objective and subjective[ii] ranges and constraints under which the data is observed and assessed. The more systematic approaches employ some form of root-cause analysis or similar trace-back methods to isolate the point of dysfunction in maintaining a data quality threshold. These methods are usually time consuming, but can be aided by the use of structured programming methods, such as UML, to develop process and function checkpoints and range verification methods. Intelligent agents, coded to search for specific events or data ranges, represent another means of dynamically screening for potentially poor data quality throughout the data fusion architecture, especially data sources.

A variety of techniques exist to confirm data model quality. Approaches are usually drawn from the software development field, in particular the verification and validation techniques[31]. More than 45 different techniques are available to verify and validate models, employing such techniques as boundary value analysis, event tree analysis, critical timing/flow analysis, sensitivity analyses, fault tree analysis, debugging tools, and others. Requirements for model verification are usually part of an agency’s overall information policies and procedures.

The final perspective listed in Figure 4-5 on data quality is based on an agency’s or owner’s information policies and practices. Some organizations do not have widespread, codified policies or procedures, only isolated references to the general use of data, acknowledgement requirements, and accessibility or security. For more progressive organizations, a minimum threshold of data quality acceptance is usually defined. This threshold has been established based on the perceived or actual business consequence of falling below the threshold, e.g., missed deadlines, legal liability, cost to operations, or unbearable customer criticism leading to the likely loss of clientele or support. The elements of the data policy and procedures are usually based on the data quality dimensions listed in Figure 4-5 and the previously mentioned evaluation and correction techniques preferred by an agency or organization. Ultimately these procedures are encapsulated in some form of continuous improvement process (CIP), such as basic TQM practices, ISO-based processes, or Software Engineering Institute (SEI) certification using the Capability Maturity Model (CMM) construct.

Ultimately, increased attention on data quality will occur after consumers and managers recognize the importance and transactional value of data quality. The availability of proven methods to correct near-term data elements and to implement long-term organizational processes and policies are not the constraints.

4.5 Section 4 References

General ATIS References

Abernethy, B. (Feb/Mar 2001). Road to Nowhere: The Ongoing Standards Saga. Traffic Technology International.

Article. The ITS industry faces being burdened by standard that are often untested and unwanted. ITS standards development process has serious shortcomings. The author recommends to return the standards process to that which proved successful 'pre-ITS'.

Albright, N. (1/25/01). 511 Case Studies in Kentucky.

Focuses on the Commonwealth of Kentucky and its implementation of statewide 511 services. Provides a concise, current "snapshot" of the progress being made. Includes sections on history/perspective; institutional background; plans/visions; ongoing activities; and lessons learned.

Allied Business Intelligence (4/25/01). Despite Slow Start, Satellite Digital Radio Industry Will Flourish, According to New Digital Car Study from ABI. Oyster Bay, NY.

news. Satellite-based digital audio radio services (SDARS) broadcasters, XM Satellite Radio and Sirius Satellite radio, will ultimately benefit from increased consumer uptake an large recurring service revenues. By 2006, recurring annual service revenues for SDARS will reach $350 million, according to the findings in "The Digital Car: A Strategic View of Global In-Vehicle Communications Technologies and Next-Generation Telematics Systems," a new study from Allied Business Intelligence (ABI).

Allied Business Intelligence Inc. (12/12/00). Wireless Internet to Drive Voice Recognition, According to New Allied Business Intelligence Study.

Discusses the pace of growth in the satellite navigation industry, particularly GPS, the first and only system that is fully operational with existing users. The US GPS system addresses a plethora of industries including agriculture, aviation, communications, in-vehicle, marine, recreation, science, surveying/mapping, and timing.

Almeroth, K., et al. (Jul/Aug 1999). An Alternative Paradigm for Scalable On-Demand Applications: Evaluating and Deploying the Interactive Multimedia Jukebox. IEEE Transactions on Knowledge and Data Engineering, Vol. 11, No. 4: 658-672.

Straightforward, one-way delivery of audio/video through television sets has existed for many decades. In the 1980s, new services like pay-per-view and video-on-demand were touted as the "killer applications" for interactive TV. However, the hype quickly died away, leaving only hard technical problems and costly systems. As an alternative, we propose a new jukebox paradigm offering flexibility in how programs are requested and scheduled for playout. The jukebox-scheduling paradigm offers flexibility ranging from complete viewer control (true video-on-demand), to complete service provider control (traditional broadcast TV). In this paper, we first describe out proposed jukebox paradigm and relate it to other on-demand paradigms. We also describe several critical research issues, including the one-to-many delivery of content, program scheduling policies, server location, and the provision of advanced services like VCR-style interactivity and advanced reservations. In addition, we present our implementation of a jukebox-based service called the Interactive Multimedia Jukebox (IMJ). The IMJ provides scheduling via the World Wide Web (WWW) and content delivery via the Multicast Backbone (MBone). For the IMJ, we present usage statistics collected during the past couple of years. Furthermore, using this data and a simulation environment, we show that jukebox systems have the potential to scale to very large numbers of viewers.

Anderson, S. M. I. (3/5/01). 32-bit Power Drives the Intelligent Car. Austin, TX.

(news) 32-bit microcontrollers are on the fast track to winning many applications within the smart vehicle. Last year, according to Strategy Analytics Inc. (London), 32-bit processors powered 10 percent of the vehicle. By 2005, 32-bit solutions are expected to take over 25 percent of the processing power in the vehicle.

ANSI and USDOT (1999). ANSI TS 286: Commercial Vehicle Credentials.

His standard was developed and is maintained by Accredited Standards Committee (ASC) X-12, Electronic Data Interchange (EDI) of the American National Standards Institute (ANSI). The standard contains the format and established the data contents of the Commercial Vehicle Credential Transaction Set (TS 286) for use within the context of an EDI environment. Operating a commercial vehicle in the US requires many credentials. Vehicles must be titled and registered. Carriers must have adequate liability insurance and must be authorized to carry certain types of cargo. Special permits are required to operate vehicles that are over the legal weight of size. Drivers must be licensed to drive whatever size vehicles they intend to operate, and must meet medical standards. Carriers must pay fuel taxes for operating vehicles in each jurisdiction. Some states have additional credential in requirements. This bi-directional transaction set can also be used by authorizing jurisdictions to transmit electronically credential data to applicants and other authorized entities.

ARC Group (4/12/01). In-Car Telematics Fitted in 56 Million Vehicles by 2005.

According to a recent study by ARC Group, the automotive telematics market is expected to undergo an average growth rate approaching 90% a year over the next five years, bringing the total number of cars fitted with telematics systems to 56 worldwide by 2005 compared with just over one million today.

ATX Technologies Inc. Emergency Services.

(news) ATX provides emergency services including: automatic collision notification; emergency response; basic roadside assistance; enhanced roadside assistance; remote door unlock; locator service; stolen vehicle tracking; and security system notification.

ATX Technologies Inc. Information Services.

(news) ATX Technology Inc., provides information services including: messaging services, vehicle operation information; and information services (weather, financial and sports for location-enhanced information).

ATX Technologies Inc. Motorist Demand for Telematics: Increased Mobility Increases Demand for Telematics.

(news) The growing demand for telematics, or location-based emergency, navigation and information services has market analysts forecasting a subscriber base of 1 million by 2004.

ATX Technologies Inc. Navigation Services.

(news) ATX Technologies Inc., provides navigation services including: emergency navigation; connected navigation; traffic information; and dynamic road guidance.

ATX Technologies Inc. (2/21/01). ATX Technologies Company Information.

Provider of vehicle and wireless device applications (emergency services, navigation services, information services) and resource operations (response center technology, skilled response specialists, and working with police/911)

ATX Technologies Inc. (3/15/01). ATX Technologies Business Growth to Result in Call Center Expansion.

To meet the demands of a rapidly developing market for telematics, ATX announced plans to open a second voice and data interaction center in the central or southeastern US.

Baca, M. e. (1998). Metadata: Pathways to Digital Information, Getty Information Institute: i-iv, 1-41.

A collection of essays on metadata* (*treated as plural), particularly for the World Wide Web. Gilliland-Swetland presents an overview, outlining types, functions, attributes, and characteristics of metadata, with examples from the "real world." Her essay demonstrates the importance and role of metadata in the current information universe. Gill's essay focuses on metadata in the context of the World Wide Web, and examines three important emerging Web metadata standards. Cromwell-Kessler discusses the importance of mapping different metadata standards to facilitate interoperability, and identifies some f the concomitant issues, benefits, and necessary future steps.

Barry, C. and M. Lang (Apr/Jun 2001). A Survey of Multimedia and Web Development Techniques and Methodology Usage. IEEE, National University of Ireland, Galway, Ireland: 52-60.

The survey results suggest that no uniform approach exists to multimedia systems development and that practitioners aren't using the multimedia models cited in the literature. Developers need new techniques that capture requirements and integrate them within a systems development framework.

Belo Interactive, I. (4/4/01). Belo Interactive Announces Launch of 'My Traffic': Powered by Strategy.com, My Traffic Keeps Consumers Informed of the Traffic Conditions That Affect Them. Dallas and Vienna, VA.

PRNEWS. Belo Interactive, Belo's Internet subsidiary, and Strategy.com™ Incorporated, a provider of one-to-one messaging through Web, wireless and voice, announced the launch of My Traffic on March 30, 2001. My Traffic is a personalized service that provides traffic conditions for personally selected routes to the desktop, email, or to wireless devices. In addition, My Traffic immediately alerts subscribers via the Web, email or wireless devices of problematic traffic conditions and offers information on alternative routes.

BeVocal (1/17/01). BeVocal Company Information.

Created voice portal applications that can be based on a caller's location, delivered to any device, and customized via any platform. Voice Portal applications include nationwide business finder, driving directions, and traffic updates and worldwide weather.

Blackwell, D. (3/18/01). Traffic Service to be Launched.

Global Telematics, which supplies satellite-based tracking systems for vehicle fleets, will announce a service giving up-to-the-minute news on traffic congestion, roadwork, and other delays to its users.

Boll, S. and W. Klas (May/Jun 2001). Z_YX - A Multimedia Document Model for Reuse and Adaptation of Multimedia Content. IEEE Transactions on Knowledge and Data Engineering, Vol. 13, No. 3, IEEE: 1041-4347.

"Advanced multimedia applications require adequate support for the modeling of multimedia content by multimedia document models. More and more this support calls for not only the adequate modeling of the temporal and spatial course of a multimedia presentation and its interactions, but also for the partial reuse of multimedia documents and adaptation to a given user content. However, our thorough investigation of existing standards of multimedia document models such as HTML, MHEG, SMIL, and HyTime leads to us the conclusion that these standard models do not provide sufficient modeling support for reuse and adaptation. Therefore, we propose a new approach for the modeling of adaptable and reusable multimedia content, the Z_YX model. The model offers primitives that provide - beyond the more or less common primitives for temporal, spatial, and interaction modeling - a variform support for reuse of structure and layout of document fragments and for the adaptation of the content and its presentation to the user context. We present the model in detail and illustrate the application and effectiveness of these concepts by samples taken from our Cardio-OP application in the domain of cardiac surgery. With the Z_YX model, we developed a comprehensive means for advanced multimedia content creation: support for template-driven authoring of multimedia context and support for flexible, dynamic composition of multimedia documents customized to the user's local context and needs. The approach significantly impacts and supports the authoring process in terms of methodology and economic aspects."

Booz Allen Hamilton (9/1/98). ATIS Field Operational Test Cross-Cutting Study (ITS FOT): Advanced Traveler Information Systems. McLean, VA.

Summarizes and interprets results of several FOTS that have traveler information components. Analysis and results are categorized as impacts, user response, technical lessons learned, institutional challenges and resolutions, and cost to implement. Highlights successes and problems these tests have encountered while attempting to develop the technologies appropriate to establishing and implementing ATIS.

Business Wire About Cox Radio/ About Traffic.com.

Cox Radio is the fourth largest radio company in the US based upon net revenues. Cox Radio will own, operate or provide sales and marketing services for 83 stations clustered in 18 markets, including major cities such as Tampa, Miami, Orlando, Fla., Atlanta, Houston and San Antonio. Traffic.com is creating the premier traffic information franchise with its exclusive TrafficPulse digital sensor network, and Advanced Traffic Information Services (ATIS). The network continually measures traffic flow on major highways to provide motorists with real-time actual speeds and point-to-point travel times.

Business Wire (2/28/01). Traffic.com to Provide Content for Traffic Reports on Cox Radio's Six Tampa Stations: Listeners to Benefit form More Accurate Traffic Reports. Wayne, PA.

news. Traffic.com, a provider of digital traffic and logistics information announced that beginning March 1, it will supply the data for traffic reports on all six Cox Radio stations in Tampa, Fla., the nation's 21st largest radio revenue market.

Canadian Corporate (2/22//01). WebTech Wireless Launches End-to-End Wireless Vehicle Location System for GSM Operators. Cannes, France.

(news) WebTech Wireless Inc., a global vehicle tracking and location services provider, announced the launch of the Quadrant Vehicle Location System™. The Quadrant System is the industry's first end-to-end, wireless vehicle location system that delivers a suite of location services on Palm Powered™ devices, such as Palm™ and Handspring handheld computers (PDAs), through an online services portal. The Quadrant Vehicle Location System enables GSM network operators to realize new data service revenue streams with commercial fleet and vehicle location services.

Canadian Corporate News (2/22/01). OFDM Forum First Anniversary Highlighted by Promise of Future Use of OFDM in Vehicle-to-Vehicle Communications. San Francisco, CA.

(news) The OFDM Forum, an association organized to promote a single standard for high-speed wireless communications conclude its first meeting of the 2001 calendar year. The achievement that demonstrate the importance of OFDM technology to the future of the wireless industry, as well as the need for a single compatible global standard for OFDM technology are: a demonstration of fully automated automobile that will use OFDM technology; the addition of 11 new members, including 3Com and Runcom; and the review and discussion of various proposals by OFDM Forum Working Groups.

Canadian Electronics (2/1/01). E-Vehicles: 'Telematics' Influences Automotive Electronics. Detroit, MI.

1.1.1 General ATIS Model