Video Moving Target Detection And Tracking Algorithm

Moving object detection and tracking technology in video sequences is one of the important research topics in computer vision. It combines image processing, pattern recognition, automatic control, artificial intelligence and many other advanced technologies in computer field, and has been widely used in intelligent monitoring, traffic control, machine intelligence, medical diagnosis and other fields. With application requirements in complex environment increasing day by day, how to improve the robustness and accuracy of these object detection and tracking algorithms has become the current focus of the study. This thesis focuses on the research of the commonly used target detection and tracking algorithms, and does some improvements to these algorithms.In moving object detection, this thesis first introduces the basic theory of optical flow, temporal difference and background subtraction. In order to overcome the limitations of the traditional background subtraction method which requires a pre-stored background image, an object detection method based on the combination of temporal difference and background subtraction is introduced. First of all, temporal difference and the median filter methods are used to dynamically build the background model, then the foreground object is extracted by background subtraction. Experimental results show that this method can create an accurate background model and get good foreground detection results when moving object exists in the scene.In moving object tracking, this thesis focuses on the object tracking algorithm which based on Camshift. First, the basic theory of Mean Shift and Camshift algorithms is introduced, and then these two algorithms are applied to track moving objects in video sequences. Experimental results show that Camshift algorithm may lose the target when the color of the target and background are similar. To solve this problem, this thesis first established a joint color and texture histogram model of the target which is based on Local Binary Pattern(LBP) texture and Cb, Cr color components, then on the basis of the target model, an improved Camshift algorithm based on joint color and texture features is proposed. Comparative experimental results show that in strong background interference, the improved method is still able to accurately track the target and get better performance than the traditional Camshift algorithm.In this thesis, Kalman filter which based on motion estimation is also introduced and applied to track moving objects in video sequences. The experimental results show that it gets good performance in tracking. Then for tracking fast moving objects, this thesis does research on the object tracking algorithm which combines Kalman filter and Camshift together. The experimental results show that the combination of the two algorithms not only reduces the search time, but also improves the tracking accuracy.