Object Tracking Algorithm With Kernelized Correlation Filters Assisted By Feature Points

As an important branch of computer vision,visual object tracking technology has played an important role in many fields such as human-computer interaction,biomedicine,and intelligent monitoring.In recent years,the object tracking algorithm based on correlation filtering theory has become the mainstream algorithm in the field of visual object tracking because of its good robustness and high efficiency.However,there are still many key issues that need to be resolved in this type of algorithm,such as object scale scaling,rotation,complete occlusion,and object disappearance.Therefore,this article focuses on these key issues and conducts in-depth research on Kernelized Correlation Filters(KCF).The main research contents and achievements are summarized as follows:(1)A scale and rotation-angle adaptive object tracking algorithm with kernelized correlation filters(SR_KCF)is designed.In order to complete the KCF's adaptive scaling and rotation angle improvements,The algorithm combines the forward-backward optical flow method to obtain the stable feature points of the current frame,and then the object scale and rotation angle are estimated by calculating relative distances and angle changes between the stable feature points and feature points in the object model of initial frame.At the same time,the feature points that are processed by the hierarchical aggregation clustering algorithm are used to estimate the object center,the center and the object center estimated by the KCF are effectively fused to further improve the accuracy of the object center position estimation.In addition,the KCF uses a fixed learning rate to update the classifier,which can not deal with the problem of occlusion of the object part.So the SR_KCF algorithm dynamically adjusts the learning rate,object scale,and angle according to the feature point matching rate that can reflect the degree of occlusion of the object,which can improve the algorithm's anti-blocking and accuracy.Finally,the comparative experiments show that the algorithm can achieve adaptive tracking of the object scale and rotation angle,and the algorithm can track the object more accurately than the comparison algorithm in some complex scenes such as partial occlusion and motion blur.(2)A long-term tracking algorithm based on feature point matching and SR_KCF is designed.The original KCF tracks the object only in the local area of the previous frame's object position,which cannot handle long-term occlusion or disappearance of the object that may occur in long-term tracking.Therefore,this article uses the above SR_KCF algorithm as a tracker,then designs a object re-detection algorithm based on feature point matching as a detector,and combines this detector with an effective anomaly judgment strategy to assist the SR_KCF tracker in retrieving the object when the object moves out of the field of view or is occluded for a long time and reappears in the field of view,which can improve the robustness of the algorithm in long-term tracking.In addition,in order to enable the detector to reacquire the object quickly and accurately,the novel feature library(feature point set)maintenance method is used to update the object model of the detector in real time during the running of the SR_KCF tracker so that the feature points in the model are not too redundant and can preserve the recent appearance of the object and its historical changes.Finally,the comparative experiments show that the algorithm has the ability to re-detect the object in the event of object loss reproduction or long-term occlusion and other abnormal conditions.And compared with other long-term tracking algorithms,the algorithm has better overall performance in long-term tracking.