Active Target Discrimination Algorithm For Robust KCF Tracking

With the continuous development of computer technology and image processing,moving target tracking has become one of the research hotspots in the field of computer vision.In view of the challenges in target tracking,many scholars have proposed corresponding solutions,especially based on kernelized correlation filters,which have achieved remarkable results in both accuracy and speed.However,most of the existing tracking algorithms need to initialize the tracking target manually first,besides only the single target can be tracking,and when the target is lost,the tracking process cannot be reinitialized again.In view of the above problems,this paper makes an in-depth study on the classical object detection and kernelized correlation filter tracking methods,combining with image feature,target saliency,and Fourier transform amplitude knowledge,to analyze the tracking targets.Finally proposes an active target discriminant algorithm for robust KCF tracking combining with moving targets,which determines the robust candidate tracking box based on moving targets,with a view to realizing the automatic real-time tracking of multi-target long-time stability for video.The main contents of the research are as follows:(1)introducing the background and significance of the paper,the current situation of the technology of moving target detection and tracking at home and abroad,and analyzing the existing chanllenges in the technology of moving target detection and tracking.At the same time,the existing methods of tracking target initialization are summarized and analyzed.(2)The principle of the classical moving object detection algorithms,such as inter-frame difference,three-frame difference,optical flow method,correlation coefficient method,Gaussian background modeling method and Vi Be,are analyzed.Using the multi-thread optimized correlation coefficient method to detect the moving object,the results of the moving object detection in the continuous frame are analyzed,and the target to be tracked in the scene is preliminarily determined.(3)To study the principles of KCF and its improved algorithms such as CN,DSST,Staple and ECO based on color and scale,as well as ATOM combining depth features,and to use one or more target of video sequence for tracking experiments to evaluate their performance,improve the tracking framework of KCF,Staple and ECO-HC to adapt to multi-target tracking,and integrate it with the target detection module to achieve the backbone of automatic target tracking about robust KCF.(4)Using target image feature,target saliency detection and Fourier transform amplitude to measure the moving object characteristics of tracking,filtering with RANSAC,and selecting a stable and reliable block for tracking.