Research On Abnormal Human Behavior Detection In Surveillance Videos

This paper is mainly focused on the research of intelligent surveillance video analysis and aimed on detecting human abnormal behaviors. Firstly, the research and development of intelligent video surveillance system as well as the applications of computer vision and artificial intelligence techniques in video analysis at home and abroad are introduced in detail. Then the author has discussed and analyzed with the usual public video datasets and the denoising techniques for video and image processing, which has laid a solid foundation for the next step.Secondly, an improved background model of Mixture of Gaussians is proposed. This method for foreground extraction is able to initialize the Gaussian mixture model quickly and adapt to rapid changes in the background, and then the weights of different Gaussian models are updated by the occurred frequency within the time window. Experimental results on different video sequences show that the method is capable to obtain most of the foreground both in indoor and outdoor environments.Thirdly, the foreground mask is labeled as several regions of interest, and the optical flow features in each labeled region is calculated using the Lucas-Kanade algorithm. The author defines the amplitude-based weighted orientation histogram to measure the anomaly of human activity and to compute the entropy of each labeled region to recognize abnormal events. The test results under different behaviors dataset demonstrate the effectiveness of this method.Finally, an intelligent surveillance video analysis software platform is established using the libraries of two open-source projects, wxWidgets and OpenCV. Then parts of the image processing algorithms in MATLAB are transplanted to the new platform and treated in modules. The new platform can be used in real-time video analysis and processing.