Improvement And Implementation Of TLD Target Tracking Algorithm In Complex Environments

Target tracking technology is an important research topic in the field of computer vision.Its core is how to accurately determine the specific location of the target in each picture sequence.Tracking-Learning-Detection(TLD)target tracking algorithm is a combination of tracking,detection and online learning algorithm,which can achieve long-term stable tracking of the target.However,TLD target tracking algorithm is difficult to achieve the ideal tracking effect in some complicated environments such as target occlusion,illumination change,scale variation and attiitude variation.In order to improve the tracking performance of TLD target tracking algorithm,the following work has been done in this paper:(1)Improvement of the tracking module.For the problem of insufficient representation effect of feature points sampled by tracking module in TLD algorithm,Oriented FAST and Rotated BRIEF(ORB)feature points are used to express the target,and to supplement the feature points by random sampling feature points to enhance the expression ability of the target and improve the stability of the algorithm in the tracking process.(2)Improvement of the detection module.For the problem of high time complexity of TLD algorithm detection module and insufficient classification accuracy of classifier,Binary Normed Gradients(BING)algorithm is used to detect and delete candidate targets that are not qualified.And then the variance classifier,Extreme Learning Machine(ELM)and the cascade of nearest neighbor classifier are used to realize target classification and detection,and to improve the detection efficiency and real-time performance.(3)Implementation and experiments analysis of the improved algorithm.For the improvement of tracking module and detection module,this paper integrates the two improvement strategies and compares them with other tracking algorithms on test sets with different characteristics.The performance of the improved algorithm is evaluated from the center position error,tracking overlap rate,tracking success rate and the running speed of the algorithm.