Surveilance Video Event Detection Algorithm In Complex Scence

Intelligent video surveillance system has been paid more and more attention by the public in recent years. It plays an important role in the field of public security. Event detection is the core of intelligent video surveillance system. The main techniques include target recognition and detection, target tracking, behavior analysis, event recognition and detection in complex scene. In this paper we focus on three aspects:pedestrian detection, multi-target tracking and event detection.In this thesis, several state-of-art pedestrian detection algorithms are studied and a detailed analysis of pedestrian detection method based on Edgelet feature is given. We improved it by proposing a novel feature called Edgelet-LBP. A weak learning method based on feature space partition is used in training stage of Real Adaboost detectors, which matches our feature well. Experimental results show that the proposed algorithm is comparable with the state-of-art methods. The performance of head-shoulder detector is largely improved, which is essential in dealing with partial occlusions.A multi-target tracking algorithm based on generalized minimum clique problem is implemented and improved in this thesis. The algorithm considers both global and local features of appearance and motion of the targets, which is the basis of a better performance of multi-target tracking. Experimental results show that our modifications accelerate the running speed and maintain the tracking accuracy in video segments as well.Two event detection algorithms are implemented. One is based on space partition and the other one is based on pedestrian detection and tracking, which are used in2012and2013TRECVID SED evaluation respectively. Valuable experience and lessons are accumulated. We compared the two methods and provided a detailed conclusion.