Design And Implementation Of Object Tracking System Based On Correlation Filter

Object tracking is one of the main research topics of computer vision.In recent years,this technology has been widely used in many aspects such as medical treatment and technology.Due to the complex application scenarios,factors such as occlusion,deformation,rotation,and scale changes bring many challenges to the tracking algorithm.As an important branch of object tracking,the speed advantage of correlation filtering tracking algorithm has been favored by both academia and industry.Aiming at common problems of object tracking,based on the Correlation filtering theory,the thesis does some targeted work on the deficiencies of Kernelized Correlation Filter(KCF)and STRCF(Spatial-temporal Regularization Correlation Filter).(1)When the target is occluded and deformed,the KCF algorithm is prone to tracking drift problems.This thesis improves the KCF algorithm feature extraction and model update module,and realizes a multi-feature fusion anti-occlusion kernel filter tracking algorithm.First,the method of fusing Histogram of Oriented Gradient(HOG)features and Color Name(CN)features is used to improve the problem of insufficient single feature characterization ability.Second,a high-confidence model update strategy is used to determine whether to update the model.This method can improve the problem of introducing incorrect background information when updating the model frame by frame.Experiments show that the tracking accuracy of the algorithm in this thesis is higher than the comparison algorithm,and it has higher robustness in dealing with target occlusion and deformation.(2)When the target rotates in the plane,the STRCF algorithm is prone to the problem of inaccurate target positioning.This thesis improves the STRCF algorithm and implements a related filtering tracking algorithm based on spatio-temporal regularity and feature fusion.This method uses HOG,CN,and Multi-Block Local Binary Patterns(MB-LBP)features to improve the algorithm's inaccurate positioning when facing target rotation,and selects the scale pool method to estimate the target scale change,Choose Alternating Direction Method of Multipliers(ADMM)to solve the optimization model.Experiments show that the tracking success rate of the algorithm in this thesis is higher than that of the comparison algorithm,and it has higher robustness when the target rotates.(3)In order to study the related filtering tracking algorithm more conveniently,and to observe the tracking effect more intuitively,this thesis designs and implements a object tracking system based on Matlab?GUI,whose functions include image sequence reading,tracking,pause and exit.After experimental testing,the object tracking system designed in this thesis works stably and meets the functional and performance requirements.