Correction And Restraint For Object Tracking Drift Under Complex Environment

The detection and tracking of moving object is always a hot topic around the world,which combines theoretical knowledge from various fields, such as computer version,image processing and pattern recognition. Because of defects or outside interference,traditional tracking algorithms often have accumulated error in the process of long-termtracking produce tracking-point drift. With the development of computer technology, onthe basis of meet the needs of real-time, object tracking algorithm began to pay attentionto how to improve the precision of tracking. How to effectively correct and restrain thetracking-point drift in high accuracy of object tracking becomes a hot topic of researchin the field of computer vision.In this thesis, the principle analysis mainly for the various existing object trackingalgorithm under complex environment produce drift phenomenon and put forwardcorresponding solutions, finally, a large number of experiments, to test and verify thetracking-point drift correction and inhibition effect.This thesis studies the main contents are as follows:(1) Research on the background and significance of the object tracking algorithm,cause of tracking-point drift under complicated environment, tracking-point driftcorrection under complicated environment, related to the inhibition of technology ofdomestic and foreign research present situation as well as the development trend.(2) The correction and suppression method based on the template tracking pointdrift. Mainly research the causes of the tracking based on template drift: The correctionand suppression to tracking-point drift of the multiple templates smart update algorithmbased on the characteristics of accumulation. The correction and suppression totracking-point drift of tracking algorithm based on template buffer. The correction andsuppression to tracking-point drift of tracking algorithm based on fertility templatemethod.(3) Object tracking method based on online learning. Mainly studies the basictheory of machine learning. The basic principle of tracking algorithm based on onlinelearning. The basic principle of object tracking algorithm based on a semi-supervised learning. Object tracking algorithm based on online–Boosting. The correction andsuppression to the tracking-point drift depends on P-N Learning algorithm.(4) Tracking-point drift correction and suppression method based on co-training.The principle of collaborative training framework presented in this thesis, and theprinciple of AdaBoost（adaptive Boosting）algorithm, the principle of Mean shiftalgorithm, and also the tracking algorithm based on co-training framework fortracking-point drift correction and suppression.