Research Of Visual Target Tracking Method Based On Kernelized Correlation Filter

With the rapid development of science and technology and computer vision technology,intelligent video surveillance technology in people's daily lives are widely used.In the field of computer vision,moving target tracking technology as a key technology in intelligent video surveillance,has always been one of the key research topics,its intelligent security monitoring,auxiliary driving and even unmanned automatic,industrial robots and intelligent robots and ot her high Science and technology fields have a wide range of applications.This paper focuses on the KCF(Kernel Correlation Filter)and the DSST(Discriminative Scale Space Tracker),which are the best performance of the current time-varying,Target tracking is studied in this paper.The aim of this paper is to make a breakthrough in the field of target tracking failure redetection,the introduction of random jitter factor to increase the accuracy of the algorithm,multi-scale estimation to improve the algorithm precision and the application of visual target tracking.1.A method that identify object tracking failure and redetect based on PSR(peak sidelobe rate)is proposed to overcome the problem of tracking effect detection in traditional Kernel Correlated Tracker(KCF)tracking process.In this paper,the learning samples of the kernel correlation algorithm are preprocessed,and the function of the weight distribution is taken by the two-dimensional Gaussian function.The image block is weighted at the target position in the sample,so as to obtain a higher target signal to noise ratio As well as less background interference information.In the process of image sequence processing,the peak sidelobe ratio of the correlation operation is detected in real time,and it is judged whether or not the tracking fails.At the same time,the detection function is updated in the sequence of the tracking effect,and the detection function of the tracker is enhanced,To restore movement target tracking.Finally,the robustness of the target tracking algorithm is improved by PSR re-test,which is 13.2% higher than that of the traditional KCF target tracking algorithm.2.According to the random jitter factor introduced by particle filter,it is helpful to improve the stability of the tracking system.Based on the relevant filter classifier,a visual target tracking method with motion state estimation and target scale estimation is proposed.This method combines the particle filter and the kernel correlation filtering method to estimate the position of the moving target first.And then perform the scale correlation filter to estimate the target scale,which can make the algorithm have more adaptability to the moving target of scale change.The experiment proves the efficiency of the algorithm,and it has strong adaptability in the complex cases such as target scale change,illumination change,attitude change,partial occlusion,rotation and fast movement.3.Aiming at the traditional kernel-related target tracking method,the improved visual target tracking method of nuclear correlation multi-scale estimation is proposed,which improves the kernel-related tracking algorithm to the moving target with obvious scale change With higher tracking performance.In this paper,the gray angle HOG pyramid is mapped into the one-dimensional vector,and the HOG pyramid is used as the one-dimensional input,and the highest response value of the correlation filter can be obtained,and then the optimal response value is taken as the target scale.The experiment proves the timeliness of the algorithm,and verifies that the method has strong adaptability under complex scene,and has important theoretical and applied significance.4.Active vision system is more intelligent and more in line with the needs of human daily life,this paper presents a PTZ camera control method based on the proposed target tracking algorithm with motion state estimation and multi-scale estimation.The OPENCV library implements the PTZ camera control system with the engineering language C ++(on the QT5 platform).And through field experiments to verify the actual life scenes,the system can accurately track a given target and can accurately control the PTZ camera rotation,to ensure that the target in the camera center of vision.