Research On Compressive Sensing Based Method Of Image Online Learning Tracking

Target tracking is an important research field in computer vision, and gets widely used in many areas, such as intelligent surveillance, intelligent transportation, human-computer interaction, so there is great value and practical significance of doing tracking research. Since the target appearance could not be predicted during the campaign, tracking algorithm which can update appearance models become famous by online learning.In this paper, we do some researches and explorations about online learning, focusing on a long-time target tracking with online learning algorithm, and propose improvements of real-time performance on the basis of this algorithm. The main work is as follows:Do Research on several common online learning tracking methods, focusing on the TLD algorithm, which is characterized by achieving a long time tracking. For the deficiency of the real time tracking in TLD, we study the fast compressing sensing algorithm, proposed a new on-line learning tracking method CS-TLD. It's based on the theory of fast compressing sensing to track object, making it fast, efficient and timely. Compared with the previous algorithm in TLD, the fast compressing sensing algorithm has promoted the performance of real time object tracking. By studying CS-TLD algorithm, the detection module needs to detect all child windows in efficiencies in time. The method based on Kalman filter estimates target general area in the current frame, and detects the target in this general area. Although there is some loss accuracy during tracking, but the algorithm has been greatly improved in real-time performance. Implementation of the Software interface platform, analysis of the experiments results, and preparation of the simulation sequence.