Adaptive Weighted Compressive Tracking Combined With Motion Vector And Background Information

Object tracking has been a very active research direction in the field of machine vision.It has been widely used in the field of human computer interaction,video surveillance,vehicle tracking and so on.Compressed sensing tracking algorithm based on compressed sensing theory,The algorithms has received the widespread attention because of its high tracking robustness,fast calculation speed,can satisfy the practical application.In this paper,the main work of compressed sensing tracking algorithm is reflected in the following two aspects.First,in the compressed sensing tracking algorithm,if the target motion information can be effectively analyzed,It will play a key role in the prediction of the target position,and then improve the accuracy of target tracking.When selecting the target feature to train the classifier,if we filter the extracted features,choosing the features of the positive and negative samples to train the classifier will greatly increase the robustness of the classifier.Second,tracking process is often accompanied by the target occlusion,if we add occlusion detection mechanism to the algorithm,When the target is detected block,stop the classifier to continue learning can avoid Classifier occurrence wrong classification,at this point,making full use of the local background information of the target to achieve the tracking of the target after occlusion.To reduce the drift phenomenon in object tracking,a candidate object location search method is proposed combining motion vector with super pixel.In order to weaken the influence of complex background and improve the tracking robustness,the features from the blocks in the tracking box are assigned different weights according to their locations.The classifier may get wrong information if it continues learning when the tracking object is largely occluded.A object detection approach is proposed to avoid the false classification in the situations of object occlusion.The experiment results show that the proposed algorithm has better performance and can track successfully and efficiently for a long time,compared with some state-of-the-art works in many complicated situations,e.g.swift movement,object deformation,complex background,occlusion and illumination variation.To reduce the interference of background information around the target,a target block feature extraction method is proposed.The features from the blocks in the tracking box are assigned different weights according to their locations to weaken the influence of background.In order to improve the robustness of the classifier,features with better discrimination are adaptively chosen to train the classifier by using Bhattacharyya distance of the probability distribution of positive and negative samples.The classifier may get wrong information if it continues learning when the tracking object is largely occluded,so a target occlusion detection approach is proposed to track successfully when occlusion occur,which use target and local background information.The experiment results show that the proposed algorithm has better performance and can track successfully and efficiently for a long time in many complicated situations,e.g.swift movement,object deformation,complex background,and occlusion and illumination variation.The results show that the proposed algorithm has better performance and can track successfully and efficiently for a long time in many complicated situations,e.g.swift movement,object deformation,complex background,occlusion and illumination variation.