TLD Video Object Tracking Algorithm Based On Feature Point Key

Video object tracking,as frontier technology in today's society,plays an important effect in the field of computer vision,after years of development,already has a wide application prospect in national defense,medical,artificial intelligence,video monitoring,etc.The traditional The traditional target tracking algorithm due to the lack of online learning function is based on as much as possible to improve the single track of time and accuracy,once the target deformation and occlusion due to target tracking lost,will not be able to capture and track the target.TLD(tracking-learning-detection)which invented by Dr.Kalal Zdenek is a new single target long-term tracking algorithm,and it's very effective.The algorithm requires less prior knowledge can achieve long-term tracking of targets.But TLD algorithm also has some insufficiencies,in the video of target tracking,when tracked target encounter obstacles occlusion,there will be tracking drift phenomenon in the short term,in order to overcome this shortcoming,this paper use the key feature points of the TLD algorithm is improved,the increase in the TLD track detector module,the improved algorithm inhibition of tracking drift,improve the tracking speed,has remarkable effect.This paper mainly discussed the related research on long-term target tracking algorithm for TLD.Firstly,introduces several typical image feature detection methods,including Harris,Surf,Fast,Shi-Tomasi,performance between different detection method is analyzed through the experimental comparison.Secondly,introduces Pyramid LK optical flow tracking algorithm and random forest feature detection,these are the key technical links in TLD algorithm.Then,introduces the system structure of the TLD algorithm,focuses on the analysis of the working principle of each module.Finally,the TLD algorithm for the existence of tracking drift phenomenon in when occlusion,proposed in the TLD tracking module is added before the detector,Respectively using Harris feature and Shi-Tomasi feature points to replace the original Grid sampling points.Compared with the original TLD algorithm,the results showed that,two methods for the detection of feature points has significant effect,to suppress tracking drift.but,an improved algorithm of Harris feature points in speed than the original algorithm based on,in conclusion,the TLD algorithm based on Shi-Tomasi feature points has very good results in the inhibition of tracking drift,improve the tracking speed.