Research On Optimized KCF Algorithm Based On Harris Corner Detection

With the deepening of computer vision research,the tracking of moving targets is gradually becoming an important topic.In recent years,the tracking method of correlation filter has gradually become a research hotspot because of its speed advantage,which pushes the target tracking technology to a new height.But in the actual detection scene,objects are prone to occlusion,scale variation,fast movement,deformation,etc.,and the Kernel Correlation Filtering(KCF)algorithm is easily affected by the actual detection situation,which leads to the deviation of tracking and reduces the tracking effect greatly.Therefore,how to make the algorithm take into account both accuracy and effectiveness still remains to be further studied.In order to make the tracking result more accurate,the thesis proposes an optimized KCF algorithm based on Harris corner detection,and verifies its precision and success rate by simulation.Experimental results show that the proposed method can track common targets effectively and has the advantage of accurate estimation.The specific innovations of this thesis are as follows:(1)In view of the fact that KCF algorithm is susceptible to the movement of the target itself,the generalized Hough algorithm is introduced to achieve the purpose of mutual restriction and mutual correction.In view of the fact that the generalized Hough algorithm has so many edge points and is affected easily by illumination,the improvements are as follows:Using adaptive Harris corner points,which are suitable and robust can solve the problems that the generalized Hough algorithm is slow in extracting redundant edge points and incomplete in extracting edge points caused by illumination changes.At the same time,the introduction of adaptive threshold method minimizes the influence of noise on corner extraction.(2)Aiming at the shortcomings of KCF algorithm while facing target occlusion and target scale variation because of the movement from far to near,in order to make the tracking effect more ideal,the corresponding improvements are as follows: Divide the target into blocks and track each target sub-block respectively to solve the problem that the target is easy to lose when its scale changes by the relative position between sub-blocks.In addition,the learning rate parameter is updated adaptively,which reduces the learning rate of KCF algorithm and reduces the model updating error when the target is occluded.At the same time,the HOG feature of the target extracted by standard KCF algorithm is fused with the color feature of the target.Calculate the maximum value of the response function and at this time,the position of the target can be obtained,which reduces the influence of deformation on the tracking result greatly.(3)Aiming at the problem of poor tracking effect of fast moving targets,the improved generalized Hough algorithm is better integrated with the optimized KCF algorithm by using intelligent algorithm.In this thesis,multiple targets are associated with each other by Hungarian algorithm and intersection over union.Corresponding coordinates obtained by the two methods are taken out one by one.Correct them with each other,and the final position can be obtained by the target contour depicted by generalized Hough algorithm,which suppresses the drift phenomenon of KCF algorithm when the target moves quickly.Experimental results show that the proposed method can meet the time requirement of multi-target tracking and improve the reliability of target tracking effectively.