Object Detection And Tracking Technology For Video Surveillance

With the increasing scale of video surveillance systems, it is impossible to monitor all the videos manually. Therefore, intelligent video surveillance system with independent analysis ability will orientate the trend of the future development of video surveillance system. In addition, many space such as public places and enterprises also begin to promote unattended video surveillance which improving the development of intelligent monitoring.Several typical motion detection algorithms are analyzed in this thesis, and a motion detection algorithm of Gaussian mixture model combined with noise estimation is proposed according to the requirements of actual control systems. The algorithm selected the initialization parameters and learning rates of the Gaussian models adaptively based on the noise estimation results. The experimental results of this thesis show that the algorithm can increase the modeling speed of background model, reduce the false detection rate and detect the moving objects quickly and efficiently. The algorithm meets the needs of real-time tracking well.This thesis gave the analysis and comparison of typical tracking algorithms and focuses on the tracking algorithm based on particle filter. Some improvements were made to overcome the defects of particle filter. Firstly, an improvement of resampling algorithm by adding a tiny Gaussian perturbation was proposed in order to inhibit the sample impoverishment; secondly, based on the combination of the motion detection and the predictive ability of particle filter, the tracking algorithm reinitialized after a certain time for the following tracking, which ensured the diversity of particles and the effectiveness of real-time track; lastly, an adjustment mechanism of target rectangle was applied to decrease the inclusion of the background pixels. The experimental results show that the algorithm in this thesis can track the target well in complex environmental conditions, and the accuracy of the algorithm is better than the normal particle filter.In the end of this thesis, based on motion detection and tracking algorithm, designed a practical scheme of the detection of left suspicious packages by using camera feedback and preset point mechanisms, then gives the experimental results and data with corresponding analysis.