| Summary: | 碩士 === 國立成功大學 === 電腦與通信工程研究所 === 94 === In recent years, public safety becomes more and more important issue in our daily life. In order to reduce the damage from traffic incident, video-based surveillance system has a wide range of applications for traffic monitoring. In this thesis, an automatic traffic surveillance system is presented to track moving objects and detect the collision event. First, initial background model is built to extract the foreground regions by background subtraction and background model must be updated by the result of extracted foreground region due to illumination change of sunlight. Next, self-shadows are removed based on the information of brightness and chromaticity distortions between foreground and background pixels. Then, centroid matching strategy is used to find out the correspondence of regions in the successive frames, and the system not only can track individual moving objects, but also can discriminate the mergence and split. Final, the system can detect the collision event and troublemaker fleeing from traffic incident by proposed method.
|
|---|
