Intelligent Surveillance Based On Technology For Abnormal Human Behavior Identification

With the rapid development of information technology since late last century and prompted by a series of terrorist attacks on some Western countries early this century, video surveillance technology has witnessed a fast growth. In some important public places such as subways, railway stations, squares, airports, banks etc. more and more surveillance cameras have been installed to enhance security and municipal management. It is reported that in England alone, more than 4.2 million such cameras have been installed in buildings, stores, streets and stations, still more installations are under way. In our country, with the approaching of such great events as the Beijing Olympics and Shanghai World Expo and with such security projects as "the safe city" going on, the number of surveillance cameras in cities is growing exponentially.Traditional surveillance systems have two limitations. One is their heavy dependence on the humans for analysis and decision making. Limited by human physiological conditions, people can only watch over a limited number of surveillance screens for a limited period of time. The other limitation is their insufficiency of realtimeness. In many cases, the video footage can be analyzed retrospectively in playback only for evidential purpose rather than for real time alert. Therefore, a crucial problem to be solved is how to enhance the smartness and realtimeness of the surveillance system.Starting from the traditional video image processing approaches and centered on the detection of pedestrian's abnormalities in the monitored scene, the paper researches into the key algorithms for smart surveillance technology as well as system realization. This paper contributes mainly in the following three aspects: 1) Proposing three ways to realize detection of abnormal behavior under various circumstances and using them to solve several commonly seen surveillance problems in terms of abnormal behavior detection; 2) Applying the machine learning method to the smart surveillance system, thus enhancing its smartness and realtimeness; 3) Designing and realizing a multi-mode smart video surveillance system using audio signals and multiple sensor signal auxiliary videos. The demonstration and system formulation present the validity of the approaches in this paper.