A method for automatic detection of traffic incidents using neural networks

One of the most important aspects of traffic management systems is their ability to detect traffic incidents such as accidents, disabled vehicles, and obstacles on the road. The incidents affect highway drivers and cause traffic congestion, so an immediate and automatic detection method is desired. We think that the changes in traffic average in case of traffic incidents have certain patterns different from the normal case. Our research tries to detect traffic incidents immediately and automatically by using neural networks, which use one minute average traffic data as input, and decide whether an incident has occurred or not. To train the network we used traffic data from various locations where accidents had occurred and not. The former are generated by a micro simulator and the latter are collected by using ultrasonic vehicle detectors. To reduce the number of false detections so as to improve the process of training of the neural network, we added some data with similar average change patterns as observed when incidents occurred. As a result, we confirmed that adding enough combinations of similar average change patterns was very effective in increasing the recognition rate and to reduce the number of false detections.