The Research Of Active Visual Tracking Technology Based On SURF

Vision is an important way for people to get information from the world, so the goal in the research of Intelligent Machines is to make the computer have visual system. On this basis, the computer vision subject has been developed grounded on the combination of machine vision and biological vision. Human own active vision which simulates the initiative of human vision and controls the movement of the camera to gain the concerned image from a suitable angle on request the current visual requirements. Active vision research in this paper is an important research topic of computer vision tracking technology. It can be widely used in robot vision, virtual reality, intelligent control, human-computer interaction and other fields. As the use of a PTZ camera in the active vision system, it can adapt to the dynamic changes in the environment and has a greater range and flexibility compared with passive vision.Based on the former research achievements, this thesis will give a thorough study of the tracking technology for active vision. By deeply analyzing the key issues faced by the current research of active vision, this thesis also puts forward some improvements measures.As for moving object detection, the research in this thesis is based on static scenes, that is, in the case of stationary cameras to detect and avoid detection for dynamic scene brought all motion compensation problems (too computationally intensive), by which it tells the advantages and disadvantages of the frame difference method, background subtraction algorithm, optical flow method and the mixture Gaussian background model method. Gaussian mixture background model method can detect the moving objects more completely, and adapt to the outside scene changes steadily. And the complexity of the detection time can meet the real-time requirements of active vision tracking system. Therefore, the mixture Gaussian background model method is used for the moving object detection, and the morphological motion filter is used to fill the empty area and remove small background noise.As for the moving object tracking, the template matching algorithm, Mean Shift tracking algorithm and SURF feature matching algorithm will be analyzed. SURF feature matching method can minimize the impact of scene illumination, the target scale, rotation but slow in speed, so it is difficult to meet real-time tracking requirements. Hence, the suggestions on the improvements of SURF feature matching method will be given in this thesis. The scale space of the pyramid model and the method of extracting feature points have to be improved respectively. What's more, the dimension of feature vector is reduced by principal component analysis (PCA). In the tracking process, the Kalman filter is used to resolve shelter issue. The experimental results reveal that the object tracking algorithm based on Kalman filtering method and improved SURF has good real-time and robustness performance, which achieves the desired effect.Finally, the tracking technology for active vision is applied in the mobile intelligent terminals, brodadening the usage scope of the tracking technology for active vision, which is an attempt in the application.