Research On Intelligent Video Analysis Technology Under Complex Background And Implementation

In recent years, with the socio-economic and cultural development, video surveillancesystem has been rapidly growing popularity, and research of intelligent video analysistechnology has been promoted vigorously. This paper has done some research on targetdetection, target tracking and specific scene rules determination under complex background,and proposes some new methods and ideas. Here also implements an intelligent videoanalysis platform system on the basis of previous research, and the system has been verifiedin a real environment. The paper mainly contains four areas as follows:1、Detection for moving targets. For the difficulty or low accuracy on foregroundextraction in complex environment, on the one hand, this paper proposes BTP-JE model, thismodel uses Bayes criteria and the total probability formula to classify pixels as backgroundpixels and foreground pixels; on the other hand, foreground detecting by combining BTP-JEmodel and optical flow method can significantly improve the accurate of prospects forextraction.2、Determination for scene rules. In order to determine a target in the monitoring videowhether is abnormal, it is necessary to customize the rules for a scene. It is abnormalbehavior when a target is into a virtual alarm region in video surveillance. For the uncertaintyof the selected target area, this paper gets a more practical model of polygonal determinationfor abnormal target through using some relative geometric algorithms.3、Track for moving targets. CamShift algorithm may be unable to track the same objectcontinuously when target is moving away from video surveillance area or part of the target iscovered, to solve this problem, here proposes an object tracking method which is based onCamShift algorithm and optical algorithm, and problem of target tracking loss can be solvedto some extent. CamShift is a semi-automatic tracking algorithm, which needs initializesearching area manually, and it can only track one single target at a time. Aiming at theseproblems, this paper proposes an object tracking approach based on OTBKC-BTP-JEalgorithm. The experimental results show that the algorithm in this paper can solve automaticmulti-target tracking problem well, and it has good stability too. 4、Development of intelligent video analysis platform. This paper designs and developsan intelligent video analysis platform system on the basis of the previous research, the systemcan not only realize intelligent monitor for residential quarter, road and school and otherplaces, but also can do some algorithmic test.