Research And Implementation Of Moving Object Tracking Algorithms Based On PTZ Camera

Moving object tracking based on PTZ camera which is also called active target tracking is, in each image of sequence’s images, to find the location of the interested object and control PTZ camera to rotate and zoom in real time in order to keep the moving object in the center of the field area and see the local details of the target. Active target tracking is the core of computer vision and has important practical value and broad prospects for development in military visual guidance, robot vision navigation, traffic monitoring and so on. Because of the external illumination, tracking object’s deformation, target’s block and the control of a PTZ camera, moving object tracking based on a PTZ camera become a challenging task.First, the current two methods commonly used in PTZ tracking:Mean Shift and Particle Filter were analyzed in this thesis and mainly introduced the application, the advantage and disadvantage of the two algorithms. Second, while the currently used methods for PTZ tracking only considered the target itself and the background information around the target was ignored, online learning tracking methods came into being. Among online learning tracking algorithms, the principle of tracking algorithm on the basis of AdaBoost classifier was focused on. Because the process of updating the AdaBoost classifier took a long time and the tracking process must proceed in real time, an algorithm of online updates on the AdaBoost classifier was applied to tracking which has been successfully used to identify a vehicle in real time. Experimental results showed effectiveness of the algorithm and the tracking results were better than the Mean Shift algorithm. However, the tracking algorithm based on online AdaBoost classifier could lead to the accumulation of errors and decrease of the classifier’s accuracy, which resulted in tracking box’s drift. To solve this problem, another classifier—MILBoost classifier was introduced and on the basis of MILBoost classifier, a new algorithm called Particle Filter was proposed. Experiments showed that this method was prior to tracking method based on AdaBoost classifier and at the same time, we compared the proposed algorithm to Particle Filter method based on color and gradient. Tacking results also showed that the proposed method was better. Tracking and control system based on PTZ camera was constructed in this thesis. On the one hand, the PTZ camera control protocol was analyzed and fuzzy control strategies were formulated according to the tasks and difficulties of active target tracking. On the other hand, the tracking algorithm of Particle Filter based on online MILBoost and fuzzy control strategies were applied to the system. Finally, the experiments proved the effectiveness of the system.