Incident Detection Through Video Image Processing

Michalopoulos, Panos G.; ,

Incident Detection Through Video Image Processing

by Panos G. Michalopoulos, Univ of Minnesota, United States,

Document Type: Proceeding Paper

Part of: Applications of Advanced Technologies in Transportation Engineering

Abstract:

A machine vision system for vehicle detection and traffic parameter extraction (called AUTOSCOPE) was recently developed at the University of Minnesota. Its advantages lie in its multispot wide area wireless detection capabilities and the ability to extract traffic parameters, such as density and queue length and size, that cannot easily be obtained with conventional devices. The system was extensively tested at the laboratory and later installed at several freeway and intersection locations and compared with loop detectors on a continuous 24-hour basis over a period of eighteen months since the writing of this paper. Following the necessary adjustments, its performance and reliability was demonstrated to be equal to or exceeding that of conventional devices. As a result, the device is currently being implemented at a 3.5 mile freeway section of the I-394 freeway in Minneapolis, Minnesota for automatic incident detection. To further exploit this opportunity a total of 38 cameras were installed in this section for detailed and continuous lane by lane traffic parameter extraction on the freeway and its ramps. The use of the AUTOSCOPE system for automatic incident detection is briefly presented.

Return to search