Research Of Pedestrian Abnormal Behavior Detection Method And Implementation Of The Detection System

As the awareness of security is increasing in the international community, the demanding to the effective real-time pedestrian abnormal behavior detection method is becoming more and more insistent. The existing real-time abnormal behavior detection method is limited to some simple predefined abnormal behavior. In view of this situation, this paper proposes one real-time pedestrian abnormal behavior detection method based on double track features modeling. Double track features include local track feature and global track feature. The local track feature can be acquired by detecting and tracking the feature point in the pedestrian area, and the global track feature can be acquired by aggregating all the local track features utilizing the framework of Bag of Visual Words (BoVW). The rich local track features can represent the various local actions of pedestrians, and the global track feature aggregating all the local track feature encodes the relationship among these local track features to a certain extent. This paper models these two track features respectively, and then combines these two models to implement a robust real-time pedestrian abnormal behavior detection method and experiments this new method in a practical scene. The results demonstrate the effectiveness of this method. Finally, this method is applied to the surveillance system to realize the auto detection of pedestrian abnormal behavior.The main contribution of this paper is to:(1) experiment some classical moving target detection algorithms, and then implement one pedestrian detection method based on heads and shoulders detection combining with the code book algorithm, (2) present one novel track feature description method based on BoVW to strength the ability to describe the pedestrian behavior, (3) propose one real-time pedestrian abnormal behavior detection method based on double track features modeling based on the complementarily of local track feature and global track feature and present one effective method to extract the abnormal frame sequence, (4) realize the abnormal behavior detection system and propose one storage algorithm based on event to compress the video.