Research On Target Tracking Algorithm Based On Kernel Correlation Filtering In Complex Scene

The correlation filtering algorithm determines the target position through the similarity between the motion model and the detection target.The concept of correlation filtering has been receiving widespread attention since it is applied to target tracking.Kernel correlation filter algorithm has quickly become a research hotspot because of its high speed,high accuracy and high robustness.However,due to the diversity of monitoring scenes,the difference of monitoring equipment,the target may cause inaccurate tracking or tracking failure because of obstacles occlusion,morphological change,proportion change,light intensity change and so on.Based on this,the kernel correlation filtering algorithm is improved and applied to target tracking.The main research contents are as follows:1?In order to improve the accuracy and robustness of target tracking in complex backgrounds,this paper proposes a multi-feature fusion kernel correlation filtering algorithm.First,the CN features of the target and HOG features are reduced to form a certain feature matrix.This article uses the idea of principal component analysis to extract the significant feature information of the target in real time,reconstruct the feature matrix.Finally,the fusion feature matrix is used to train the scale filter to further improve the robustness of the algorithm.The experimental results show that the improved feature algorithm has higher recognition accuracy in complex scenes,and can perform long-term stable target tracking such as background mottle?lighting changes?target occlusion and deformation.2? A new model is constructed by introducing time regularization correlation filter,and an adaptive weighted target tracking algorithm combined with time regularization is obtained.In this paper,after the introduction of time regularization,the time regularization is integrated into the KCF framework,and the continuous convolution method is added to the loss function to train the filter.After improving the SRDCF algorithm,the phenomenon of filter overfitting is better solved.Through the analysis of experimental results,the accuracy of the algorithm after the introduction of space-time regularization reaches 54.5%,and the success rate score is 61.6%.It can better solve the tracking problem under occlusion and can adaptwell to the problem of large appearance changes.The algorithm performs well interms of accuracy,robustness and speed,and can track targets in real time.