Incident detection system by sensor fusion network employing image sensors and supersonic wave

A lot of image sensors are employed for the purpose of incident detection, because image sensors can provide much more rich information than spot sensors such as supersonic wave sensors. In addition, image sensor can overlook wide area. Therefore, installation cost is lower. However, it is quite difficult to achieve high accuracy by image recognition methods because of their instability against environmental changes. And image processing requires more CPU performance. On the other hand, for example, supersonic wave sensors have advantages on robustness against environmental changes, and they require less CPU performance than image sensors. Therefore, future event detection system should combine different sensors in order to realize a totally efficient surveillance system. In this paper, we developed algorithms for incident detection by sensor fusion technique between the two different sensors. The recall rate and false alarms was evaluated by using 3 month data of images and supersonic waves on expressway in Tokyo containing about 20 incidents. And the algorithm was then proved to be more accurate than the algorithm using a single video image which we previously developed for incident detection