Object Tracking Algorithm Based On TLD Framework

As one of the important research direction in the field of computer vision,object tracking has achieved fruitful research results through long-term development and has been widely used in security monitoring,intelligent transportation,medical diagnosis,national defense military and other fields.TLD(Tracking-Learning-Detection)tracking framework,which organiclly combines local tracking and global detection,is a hot research topic in the field of object tracking in recent years.Traditional TLD algorithm uses Median-Flow tracker to track the target in local area of the image,at the same time the cascaded classification is used to detect target in the whole image,finally the result of the combined is used to determine the target location and update the model and the parameters.This algorithm has obvious advantage in long-term target tracking,but its performance in dealing with target appearance changes and occlusion problem remains to be improved.The main work is as follows:This paper expounds the structure and principle of TLD tracking framework,systematically introduces the Median-Flow tracker,cascaded detector and the P-N learning and analyzes the advantages and disadvantages of the algorithm.A compressive tracker based on the theory of layering and weights of blocking is proposed: using two complementary ways to construct two sparse matrix,each of which extract the image color(gray)and texture feature respectively;exploding the images into a series of layers and gaussian smoothing them;for each layer,dividing the image into a series of sub-blocks and calculating weights of each sub-block belongs to the goal;using the weighted Bayesian classifier to calculate probability sample,get the target location.A semi-supervised learning based on fuzzy set theory is put forward: establishing four different constraint rules from time,space,size and similarity aspects;using the fuzzy set theory to synthesize the rules and judgment the properties of every sample.On the basis of TLD framework,we use the improved compression sensing algorithm for locally target tracking and detect the target globally by using the traditional TLD detector.Finally,semi-supervised fuzzy learning is used to identify the results of the detector and update the tracker and detector.The experimental results show that the proposed method can effectively improve the tracking performance of the algorithm,the success rate is above 80%,and the tracking results is better than that of the other state-of-the-art methods.