Studies On Unusual Event Detection In Video

Video surveillance is a key means to reinforce security in some important public places such as squares, railway stations, banks, airports etc. Traditional surveillance systems depend entirely on the humans for analysis and decision making. But limited by human physiological character, people can only monitor a limited number of surveillance screens for a limited duration of time. It's difficult to construct a truly reliable safety system, and can't satisfy the demands of applications in practice. Therefore, an urgent problem to be resolved is how to enhance the effectiveness and smartness of the surveillance system.Starting from the traditional video image processing techniques and focus on the detection of unusual events in the monitored scene video, this paper studies into the key algorithms of intelligent surveillance technology. The major work and innovation of this paper are as follows:(1) Introduced the Independent Component Analysis with Reference (ICA-R) into video moving objects detection algorithm. By defining the sphered background component of training images as the reference signal, the algorithm using ICA-R deal with the separation problem of moving objects and stationary background in video images successfully. It is showed the detection results of this algorithm is robust even if the scene light has changed markedly. (2) Proposed a video unusual events detection method based on Hidden Markov model (HMM) which using features extracted by Independent Component Analysis (ICA) and HP (Hodrick-prescott) tendency filter. This method first employ ICA to construct a normal video feature subspace, and project the image sequence into this subspace to achieve data reduction. Then the trend component in feature sequence caused by the adverse environment factors is cancelled by the HP filter, so to enhance the robustness of HMM video unusual events detection at complex outdoor scene. (3) Studied a multi-granularity video unusual events detection algorithm based on infinite Hidden Markov Model (iHMM), the algorithm uses iHMM to model and analysis the image features, and combines with the consideration of the events characters in different time granular. This method can solve the problem that current unusual events detection methods didn't consider the granular differences of various events, so has higher detection performance.The unusual events detection algorithm in real scene is vulnerable to environmental conditions. In the study of this paper, not only try to enhance the detection rates of the algorithm, but also consider the interferences by various negative environmental factors of complex scene. The presented ICA-R video objects detection algorithm can deal with the impacts of the light variation. These two unusual events detection methods described in this paper, can overcome the influences to video unusual events detection caused by luminance variation, shadow and occlusion of trees etc. Experiments validate the effectiveness of these algorithms.