Research On Video Object Track Under Complex Scenes

Video object tracking is the hot topic in the field of computer vision at present, which can be widely used in the applications of smart video surveillance, human-machine interface, robot's vision, military field and so on. But because of the complexity of the problem, video object tracking is still hard in computer vision. Some of the key-technology need to be resolved, such as losing and reappearing of the object, translation, rotation, scaling, affine distortion, quickly ruleless movement. The video object matching is the core of the technology. To solve these problems, reading lots of related papers published at home and abroad, spending much time in researching the theories and applications of SIFT, Kalman filter and Mean Shift, aiming to implement real-time video object tracking in complicated surroundings, a real-time and robust video tracking algorithm to fit different situations is proposed.Shot detection and key frame extraction of the video sequence can solve the problem of the losing and reappearing of the object and real-time request of the algorithm. A block of color histogram video key-frame extraction algorithm is proposed in this paper, in which the color and content character of the video frames as well as human eye's visual character are fully considered and each video frame is divided into 5 blocks, then the similarity between the frames using different weight of different blocks is calculated. The method can solve the problem of shot detection and key-frame extraction well.In the process of video object matching, comparing the merits and shortcomings of the traditional template matching, corner matching, Hausdorff distance matching, objects'invariant moments matching, video object matching based on SIFT algorithm is proposed. It can match the objects with translation, rotation, scaling, affine distortion, different illumination, even when the object is partially sheltered. The algorithm is real-time and can be used in the real-time video object tracking system.An algorithm combining Kalman filter and Mean Shift is brought forward, aiming at the Mean Shift couldn't track the fast moving object. Kalman filter is used to forecast possible position of object in the next frame, then Mean Shift search the real position near the possible position. This algorithm has good effect to fast moving object, and can deal well to the occlusion. In the processing of object tracking, the bandwidth of kernel histogram is self-adapting changed, the template of object is updated, too. Finding object in new shot swiftly by the SIFT algorithm, tracking object by the combination of the Kalman filter and Mean Shift. The template update algorithm in this paper makes full use of the intermediate value of Kalman filter and Mean Shift, so that the complexity of the algorithm is not increased. Template update algorithm improves the robustness of original algorithm. Experimental results show the improved algorithm is real-time. The video object tracking system is implemented.The above algorithms have been realized by experiments. The experimental results show that the algorithm has the advantage of high precision, real-time. The whole system is robustness. It has a good practicality in smart video surveillance, intelligent robot, national defense and security and so on.