Research And Applications Of Object Tracking In Intelligent Video Surveillance

Moving object tracking is one of the most popular topics in computer vision, with wide application prospect in battlefield reconnaissance, intelligent video surveillance, traffic control, human-computer interaction and so on. It attracted attention from scholars and industry, with wide application value and higher theoretical significance.Currently, great efforts have been taken on object tracking from both academia and industry with lots of valuable achievements. However, on-line object tracking is still a difficult issue in the literature of pattern recognition, there exist harsh requires for robust on-line tracking algorithm which can handle changes of viewpoint, pose variation, distance, rapid motion and occlusion, a single approach can't handle all the problems. Multi_methods ensemble is an effective strategy for robust on-line tracking. In order to solve the problems, we complete the work as following in this thesis:1. Focused on several object tracking algorithms including Mean shift algorithm, on-line boosting based tracking algorithm, particle swarm optimization based tracking algorithm and template match based tracking algorithm. Related experiments have been done on these algorithms. Then we analyzed advantages and disadvantages of these algorithms. All of the work provided relevant theoretical basis for tracking algorithm which we proposed in this thesis.2. Proposed an adaptive object tracking framework which integrate on-line Boosting with normalized cross-correlation based template matching and particle swarm optimization. Among these three methods, on-line boosting is the basic tracking algorithm; Template matching is employed to effectively prevent on-line boosting from making too many wrong updates, while particle swarm optimization based tracking strategy improve the adaptability to rapid movements, fast appearance variations ,meanwhile, it provides guarantee to update of the template, they are complementary and keep a balance between stability and plasticity. The experimental results of different test sequences show that the proposed approach efficiently alleviate the dilemma between adaptability and drifting, and successfully track the object in real-time.3. This dissertation designed and completed visual tracking based robot localization. Considering specific environment, we make use of object detection technology to extract foreground. Integrated the proposed algorithms and blobs based method for tracking. Finally, we accomplished localization task by achieving coordinate system transformation based on the projective transformation matrix.