Event Detection Algorithm Of Surveillance Video

Public safety has become a hot issue, among which the intelligent video surveillance system is an important part of the field. Specific event detection is the core of intelligent video surveillance system, including target tracking, event detection and other key techniques.In this thesis, we improve a multi-target tracking algorithm based on the solution of Generalized Minimum Clique Problem, which abstracts the video tracking as a problem for finding the optimal weight subset in the graph theory. For trajectory fragments due to the missing detection, we propose a combined algorithm to solve the problem of trajectory ID drift. Compared with the original one, the improved algorithm reduces iterations of operation and improves the time performance of the system.We implement a running event detection algorithm based on the forward-backward motion history image. This algorithm locates and recognizes fast moving target through the spatiotemporal information which included in motion history image. The experimental result shows that the detection algorithm can effectively remove a large number of irrelevant information and reduce the false alarm rate.This paper also introduces the information of TrecVID SED competition, in which we present an algorithm for group events detection. The performance and defect are pointed out though the study of results, and the future work is given.