DEVELOPMENT AND FIELD DEPLOYMENT OF A MACHINE VISION BASED INCIDENT DETECTION SYSTEM

In this paper, a new automated incident detection system based on machine vision vehicle sensing via video cameras is presented. The advantage lies in the wide area detection capabilities which allow efficient detection of shock waves and other traffic parameters that cannot be easily obtained by conventional devices. In addition, video detection allows employment of ancillary information such as traffic on the shoulders, stopped vehicles, lane changing and speed differential, traffic slow downs in the opposite direction, etc. The system, installed at the Traffic Management Center (TMC) of the Minnesota Department of Transportation was previously evaluated under continuous around-the-clock operation over a four-month period beginning in December 1991, and has been running continuously since. Recent changes which have been made are: much of the processing has been moved from the system to the video detection device; new processing to detect stopped traffic (on shoulders or in lanes) has been added; incident detection algorithms have been made more robust using data collected over the last two years; the system has been interfaced to a 39-camera, 3.5-mile, live machine vision laboratory on Interstate-394 for evaluation ; and provision of video detection data to the TMC computer network for integration and archival with other detector data.