Videobased Vehicle Signature Analysis and Tracking Phase 1: Verification of Concept and Preliminary Testing

Problem

There is currently no way to automatically identify and track vehicles traveling on California freeways. With such a system, the California Department of Transportation could obtain data on vehicle flow patterns on California's freeways and arterial roadways to better manage traffic.

In order to meet this interest, a detection system must provide:

1. reliable detection and re-identification of a wide variety of vehicles under all possible traffic flow, environmental, and light conditions;
2. operate at very low power consumption to permit autonomous battery-only or photovoltaic (solor-powered) operation; and
3. be of low to moderate cost.

This report discusses Phase 1 of the research to address this problem. Phase 1 involves preliminary work to test the accuracy, reliability, and robustness of the basic technologies (a combination of video sensing, software processing of the video signal, and wireless transmission of data to other detection modules for a network of connected detection modules) upon which the detection method is based. Together, the technologies are termed Video-based Vehicle Signature Analysis and Tracking (V2SAT).

Phase 2 involves the development of experimental hardware and software for automated detection.

Phase 3 involves the design, development, and testing of a production prototype module. Based on the prototype, several such modules will be built, deployed, and tested.

Phase 4 involves the development of the wireless network components for telemetry between individual modules and local site transponders and the hardware and software components for telephone/modem communication between overcrossing transponders and a central correlation computer.

The proposed system is comprised of:

1. a detection module located on a physical structure (such as a freeway overpass) directly above each traffic lane; multiple modules make up a "site"
2. a local transponder/repeater, one per site, which receives data from up to ten modules and retransmits data via modem and telephone line to a network hub
3. a network hub, which receives the data stream from all detector modules and correlates the data to track vehicles through the monitored area

Method

The sensor is a simple video camera. The image is processed using filters and an efficient detection algorithm to develop a Vehicle Signature Vector (VSV), which is comprised of various measurements of the vehicle dimensions.

For Phase 1, two time-synchronized video systems were deployed at consecutive overcrossings spaced 0.5 to 0.6 miles apart on US Highway 101 in three different locations in the California Central Coast area. Data was collected on three different dates to cover four different illumination conditions.

Videotaped images from the field were then analyzed manually in the laboratory to test the accuracy and repeatability of the optical signature vector as a means for classifying and re-detecting vehicles. The analysis also sought to assess the general usability of the vector as a means for classifying a range of vehicles by dimensional measurements.

Findings

Under daylight conditions (midday and afternoon), an average of 98 percent of all vehicles were detected at the second site, for a sequence of 200 vehicles. For this data set and conditions, the method incorrectly matched the vehicle with the incorrect vehicle 0.87 percent of the time. Under daylight conditions, when a "reasonable time of arrival" window was used to admit only vehicles that could have traveled between sites at speed between 30 and 80 mph, the incorrect match percentage falls significantly, approaching 0 percent for the conditions of this test.

Under dusk illumination, but otherwise identical test conditions, correct matches occurred for 95.15 percent of all vehicles; false matches occurred for 2.02 percent of all vehicles in a sequence of 103 vehicles.

Under conditions of inadequate illumination (night) but otherwise identical test conditions, correct matches occurred for 75.49 percent of all vehicles while false matches occurred for 27.095 percent of all vehicles in a sequence of 102 vehicles.

It was found that supplemental illumination or use of specialized camera technologies will be required for sufficient nighttime vehicle detection and tracking.

On the basis of the results, the V2SAT method has the potential to serve as a reliable basis for non-intrusively tracking the progress of individual vehicles along an appropriately detectorized freeway network under daylight conditions.