Research On Target Tracking Algorithm Based On Kernel Correlation Filtering

In recent years,target tracking has been widely used in many fields.The research on single target tracking has made significant progress,and a large number of tracking algorithms have emerged.The tracking algorithm based on Kernelized Correlation Filter(KCF)has become one of the research hotspots.However,target tracking in complex backgrounds still faces many unsolved problems.For example,it is still a very challenging task to achieve stable and accurate target tracking when encountering background confusion,obvious illumination changes,and target occlusion.Aiming at the shortcomings of single algorithm feature extraction and inability to adapt to the scale,this paper proposes a target tracking algorithm combined with the context framework.On this basis,the target shallow features and deep features are extracted to complete the adaptive fusion,so that the algorithm has a good tracking effect.The main contents of this paper are as follows:(1)Introduce the context-aware framework to improve the classic KCF algorithm,and use the upper,lower,left,and right regions other than the target itself as an increased sampling area to improve the feature matching degree.The positioning mechanism and the scale adaptive strategy are added to ensure the accuracy of tracking.The experimental results show that the algorithm can effectively improve the shortcomings of the target under fast motion and occlusion problems,but there is still tracking drift for complex background colors and light-sensitive videos.and track failures.(2)Combine the multi-feature fusion method to improve the single problem of the KCF algorithm for target feature extraction,and complete the target feature extraction in a targeted manner,including multi-dimensional HOG,grayscale and color features.In the face of different video sequences fused with different multi-features to complete feature extraction and feature matching,the multi-feature fusion strategy can effectively detect targets and backgrounds.The experimental results show that the accuracy of the algorithm is greatly improved when tracking video sequences with complex backgrounds,but there are still The problem of target false detection and missed detection.(3)Combining the neural network for target tracking,the deep and shallow features are adaptively fused and then combined with the improved KCF algorithm to complete the target tracking.The extraction of shallow features is beneficial to target positioning,but the robustness is weak;the deep features are affected by The appearance of the target has the advantages of small influence and strong robustness,but the low spatial resolution is not suitable for precise positioning,and it is easy to cause drift.This paper combines the advantages of the two features to improve the accuracy and stability of the tracking algorithm.The tracking effect of the algorithm in this paper is observed through the experimental results.By comparing with the classical tracking algorithms BACF,Staple,SRDCF,SAMF and DSST,experiments show that the improved algorithm proposed in this paper and the multi-feature fusion algorithm has better robustness and stability,and can achieve accurate tracking of the target.Tracking has certain research significance.