AUTOMATIC INCIDENT DETECTION THROUGH VIDEO IMAGE PROCESSING
Panos G. Michalopoulos, Richard D. Jacobson, Craig A. Anderson, Thomas B. DeBruycker
Automatic Incident Detection is one of the major challenges in urban freeway operations. In spite of recent efforts worldwide, fast and reliable Automatic Incident Detection has been elusive. To a large extent this can be attributed to the limitations of existing detection devices. To overcome this problem, a new wide-area video detection system called
AUTOSCOPE was recently developed in Minnesota and was installed in the field for rigorous around-the-clock testing for over two years. As a result, AUTOSCOPE was substantially
improved, weatherised and expanded to multiple camera units. Subsequently an incident detection system was developed, based on AUTOSCOPE measurements, installed at a site in Minneaplois and evaluated under continuous around-the-clock, real-time operation for over four months. In parallel to this, a 39-camera, seven-mile, machine vision, live laboratory was designed on Interstate-394 for full deployment and validation of the incident detection system. In this paper the development and testing of the machine vision-based incident detection system is presented, along with the long-term AUTOSCOPE test results and plans for future imporvements.