Target Tracking Based On TLD Framework

Target tracking and detection are very useful in our everyday life. At present in the intersection, subway station, stadiums, office buildings and other public places and facilities are equipped with cameras. But most places still use artificial in supervisory control. And the number of the camera is far greater than the number of monitoring personnel. So there will be a large number of video be ignored, causing information inadequate used. Using computer vision for computer automatic identification detection and tracking and alarm and realizing the intelligent camera is the trend of future development. The target detection and tracking is the key to achieve this goal. Target temporary disappearance and appearance change are the challenging problems. In view of this two kinds of problems, this paper uses TLD(Tracking Learning Detection) combining tracking method and detecting method. When tracker failed, TLD can use detector results reinitializes the tracker, and meanwhile use the tracking results to enlarge training sample set, improving the precision of the detector.In order to improve the tracking accuracy, this paper puts forward the following improved tracking module. First this paper analysis current several mature tracking methods, such as optical flow method, the Mean Shift and particle filtering, etc. When a target disappears or target's appearance changes, particle filter has strong robustness, compared with other algorithms. So the particle filtering is used as a tracker alternative TLD framework based on optical flow method of the tracker, which realizes particle filter tracking method based on TLD framework and has higher accuracy than original particle filtering. According to the target appearance changing problem, this paper join patch tracking method to realize the goal of appearance change fast accurate tracking. In addition this paper also changes the detection process and realizes the TLD framework of multiple targets tracking method.System using C + + and Matlab hybrid programming, the use of CAVIR: Context Aware Vision using Image- based Active Recognition database of different video experiment, proves that the method is better than the original tracking algorithm.