Research On Specific Events Detection In Video Surveillance

In recent years, with the continuous development of computer vision technology and the rising demand of security, human behavior recognition in video is becoming one of the most important subjects. This technology has an extensive application prospect, such as: intelligent video surveillance, artificial intelligence, content-based video retrieval, etc. Due to the development of Internet technology, more and more application involves the automatic human behavior detection in video.But when it is applied in the actual situation, the influence of such factors as self shelter form,background mutations, so recognition effect is not very good.Both domestic and overseas recognition algorithm was studied in this article, and carries on the improvement, which can improve recognition rate.Respectively research from the object detection, object tracking and behavior analysis this three aspects, and then improve the three aspects respectively, specific algorithm is as follows:(1)Object detection: Due to the complexity of human motion, the target detection results by using traditional ViBe algorithm are not so satisfactory. Therefore, this paper proposes a method to deal with the target detection issue by fusing an improved Canny operator with Vibe algorithm. Specifically, the ViBe algorithm is utilized to achieve the initial foreground region of a moving object; then, the improved Canny operator is applied to extract the edge information of a moving object; finally, the extracted foreground region and edge information are fused to obtain more accurate foreground region.(2)Object tracking: Tracking-Learning-Detection is considered as a highly efficient algorithm for tracking a single target. Although this algorithm can re-track a target when the target is occluded by other targets, there still exists many shortcomings. This paper deals with the issue of target tracking by fusing kalman filter with tracking-learning-detection algorithm. Specifically, an improved Kalman filter is first utilized to enhance the reliability of tracking-learning-detection algorithm; then, the area of the target is estimated to reduce the detection region and to increase the processing speed.(3)Behavior analysis: In order to achieve the description of data dimension reduction, Fourier transform algorithm is used in this paper. It is not sensitive for frequency domain in to change the object, so it can improve the robustness of behavior analysis. Selecting optical flow field of time domain as a feature description, optical flow field for the object movement, the Angle of view, background on small changes are sensitive, which can improve the recognition rate of behaviorTo sum up, All of the algorithm and the experiment are programmed with Visual Studio 2010 on PC, mainly based on KTH database to do experiments. By analyzing the experimental results, we proved that the improved algorithm is proposed in this paper to improve the human specific behavior recognition rate in the video.