A new traffic incidemt detector by processing the traffic sounds.

A traffic incident detector processing traffic sounds was developed. This detector automatically recognizes an abnormal sound by sound processing and transmits the images before and after the abnormal sound occurrence to the traffic control center. Some of these images are expected to allow us to analyze the incident events directly. The most important function of this detector is how to discriminate the incident sounds from the other sounds with high precision. A new sound source recognition algorithm, using a neural network with non-linearity and powerful learning mechanisms, was evaluated using actual data collected in a field experiment. Current sound source recognition rate is about 90%. For the covering abstract see ITRD E114174.