Research On Key Techniques Of Intelligent Video Surveillance System

As detecting and tracking of moving objects are two key techniques of intelligent video surveillance systems, this dissertation was focused on them. Regarding moving object detecting (MOD), an improved frame-subtraction algorithm was proposed which utilizes grads information. Moreover, the mixed Gaussian model was analyzed theoretically and implemented practically for the sake of MOD. Finally, to solve the problem that shadows usually affect the MOD methods based on the mixed Gaussian model, the approaches to rejection of shadow were investigated and a novel algorithm was proposed and validated by the experiments.Regarding moving object tracking (MOT), the existing algorithms were reviewed firstly with a focus on the Kalman tracking algorithm and the Mean Shift algorithm. Considering the complexility and characteristics of video object tracking, an improved MOT algorithm was proposed by combining the Kalman tracking algorithm with the Mean Shift algorithm. The simulation results show that the algorithm is rather robust.Finally, a framework of video surveillance system has been developed by using the OpenCV (Open Source Computer Vision Library). Thanks to the DirectShow technology applied to the framework, much higher performance was obtained in capture, processing and display of the video sequences. On these bases, a prototype of video surveillance system has also been developed, which was used to demonstrate the feasibility and applicability of the algorithms proposed in this dissertation.