City of Anaheim/Caltrans/FHWA Advanced Traffic Control System Field Operational Test Evaluation: Task C Video Traffic Detection System

Art MacCarley

Executive Summary:

A technical evaluation was conducted as a subtask of the "City of Anaheim Advanced Traffic Control System Field Operational Test (FOT)", funded by the US DOT / FHWA and the California Department of Transportation. Partners for this FOT are the City of Anaheim, Odetics Inc. of Anaheim, JHK Consulting, and the California Department of Transportation. This report is the result of evaluation Task C "VTDS Evaluation", one of three evaluation tasks covering components of the FOT. The component evaluated under Task C is a video-based vehicle detection system for actuation of traffic signals at intersections. This system, referred to as the Vantage VTDS, was developed and is currently marketed by Odetics Inc. (manufacturer) as a low-cost replacement for inductive loop detectors. It utilizes video cameras mounted on existing luminaires with a view of each of four traffic approaches at an intersection. The product cost for a four-approach intersection is quoted by the manufacturer to be $15,000 for all equipment, not including installation costs.

The FOT provided for the deployment and testing of the VTDS, and the support of a comprehensive independent evaluation. During the course of the FOT, the product line was split from the originally-proposed general-purpose detection system into separate freeway monitoring and intersection signal actuation products. Only the intersection product was evaluated under this FOT. The sample VTDS unit provided by the manufacturer for evaluation was a November 1996 release of the commercial product.

The VTDS detects the presence of vehicles in "virtual detection windows" which are established in the video image during the setup procedure, duplicating the function and location of inductive loop detectors. Setup and calibration of the system in the field requires only a standard TV monitor and serial PC mouse. The user interface for the VTDS was found to be unsophisticated but effective. Two useful features are the storage of up to four detection window setup configurations, and the option for remote setup and calibration via a serial port connection.

The evaluation focused on the detection performance of the system with respect to the intended application - the detection of vehicles on intersection approaches for signal actuation purposes. Test metrics and Measures of Effectiveness (MOEs) were developed for this purpose. Deployment specifications restricted our field tests to three signalized intersections at which detection cameras were set up and operated by the manufacturer. Video-taped field data was acquired from these intersection camera feeds, accessible at the manufacturer's facility. A 12-condition video test suite, which represented a typical range of testable traffic and environmental conditions, was assembled from this data, and from video tapes provided by the manufacturer from installations in Texas and Delaware. Documentation was provided by the manufacturer on system operation and setup. Evaluation personnel received training at the manufacturer's facility on the proper setup and operation of the system, and all tests were performed in compliance with these directions.

As means for classifying all possible types of correct or incorrect detection situations, nine vehicle detection event classes and six phase actuation event classes were defined. The VTDS test unit was sourced from the video-tape test suite, and data taken by manual observation of the response of the system for each vehicle passing through the virtual detection windows as displayed on a video monitor. Data was reduced to several composite measures of performance, designed to answer practical questions of relevance to potential users of the system. All test procedures and metrics were approved by Anaheim FOT Evaluation Oversight Team (EOT), which consisted of representatives of all FOT partners.

Among the test results: 65% of all vehicles flowing through detection windows at the intersections were detected correctly, just as they would be detected by a properly working inductive loop detector. 80.9% of all A comprehensive review of published literature and product information suggested that there is a lack of evaluation standards and meaningful test data for video-based signal actuation products. This makes direct comparison of the results of the present study with results reported for similar products extremely difficult. Following their pre-release review of this report, Odetics announced that since the completion of this evaluation, they have observed findings similar to ours in their internal test program, and that both the hardware and software of the VTDS have been subsequently replaced, resulting in significant performance improvements. We have not tested this new system. vehicles flowing through detection windows were detected adequately for purposes of proper actuation of the signal phases. An average false detection and latched detection rate of 8.3% was observed. A condition-weighted average of 64.9% of all red-green transitions, and 64.0% of all green extensions were actuated correctly. Relative to all metrics, the general accuracy of the system appeared to be good under ideal lighting and light traffic conditions, but degraded at higher levels of service and conditions of transverse lighting, low light, night, and rain. We noted problems in robustly handling low vehicle-to-pavement contrast, scene artifacts such as headlight reflections and transient shadows, and electronic image artifacts such as vertical smear, which is typical of CCD (charge coupled device) video cameras.

A comprehensive review of published literature and product information suggested that there is a lack of evaluation standards and meaningful test data for video-based signal actuation products. This makes direct comparison of the results of the present study with results reported for similar products extremely difficult.