Research Of Online Target Tracking Algorithm Based On TLD

Target tracking is a hot and difficult research in the field of computer vision. It has been widely used in the human computer interaction, visual surveillance and guidance.Although many effective object tracking methods have been proposed, there are still a lot of difficulties in practical application of object tracking due to the challenging scenarios such as significant appearance change, illumination changes, motion blur,occlusion and object scale change. Target tracking technology based on online learning can track arbitrary object while training and updating classifier so that it can track the object for long term and better adapt to the changes of objects. Due to the obvious advantages of online learning tracking technology, it quickly became the research hotspot for domestic and foreign researchers. However, there are still many problems in the existing popular online target tracking algorithms such as tracking drift, bad performance of real time and so on. The research and analysis has been focused on the TLD(Tracking-Learning-Detecting) in this paper which aims to design the real time robust target tracking algorithm. This paper has been mainly completed the following several aspects work.(1) Research the existing target tracking algorithms, introduces the development of generative target tracking algorithm and the discriminative target tracking algorithm.TLD algorithm is mainly discussed, and the advantages and disadvantages of TLD target tracking algorithm are analyzed in detail.(2) TLD target tracking method based on image feature point matching is proposed.Firstly, the best tracking area can be obtained by extracting saliency region of the intialized target area, which can solve the sensitive of the target size and improve the quality of sample selection and reduce the influence of background clutter. Secondly,the target’s scale can be estimated through median flow tracker and feature points matching and the interference point can be rejected by hierarchical clustering, based on these methods, tracker drifting and out-of-plane tracking failing can be resolved effectively. Finally, a simple fast search objectives method based on adaptive scaledetector was proposed to accelerate detection speed. Experimental results demonstrate that the proposed algorithm can improve the tracking robustness of TLD target tracking method effectively and obtain good results on standard data sets.(3) The TLD target tracking based Spatio-temporal context is proposed. Due to the gray appearance model can not express the appearance model good. So in this paper,we expand the STC(Spatio-Temporal Context) target tracking algorithm by introducing the color attribute. In order to improve the tracking speed, this paper adopts the method of PCA to reduce the dimension of the adaptive selection of the representative color attributes, and improves the update mechanism of the original tracking algorithm. While the introduction of multi-scale pyramid space to estimate the scale of the current optimal target position, to prevent the accumulation of errors phenomenon with scale update. And combining the TLD framework of the detector and the learner to achieve a long time online tracking system. Good tracking results are obtained by testing the standard data set.