Research On Intelligent Video Surveillance Under Multiple Cameras And Its Application In The Cable Anti-Theft

Stolen of the electrical cables is one of the main factors that influence the stable of our society and economy. Intelligent monitoring provides a good way to prevent this effectively. This thesis addresses the problem of object detection/tracking in a single camera system and object handoff in multi-camera systems where the overlapping views exist.First, in object detection, the background subtraction-based approach has strong adaptability to process images in complex environments. Therefore, we propose an adaptive and multi-modal background subtraction approach. In particular, we update the background model by adding the survival time as a constrain based on the Gaussian mixture model-based background subtraction approach. As as result, we solve the problems that the mixture Gaussian background subtraction approach has, i.e. background model is suppressed by high frequency noise.Second, in object tracking, since the targets are often occluded by obstacles, we combine Kalman filter and Meanshift to improve performance of the traditional tracking approach. Our approach is able to track the objects with different sizes in the video and has very high adaptability in complex environments. Experimental results show our approach can run in real time and also has high robustness.The last, in Object handoff, since the field of view(Fo V) of cameras is only determined by their positions and angles and is not influenced by the environment, we propose to combine SIFT and homograph matrix to extract the field of view, and also validate our approach through experiments.Experimental results show that our approach is able to detect and track moving objects using multiple cameras, which meets the requirements of the anti-stolen applications.