Research On Target Tracking Algorithm Based On Correlation Filter

Target tracking is an important research direction in the field of computer vision.It is widely used in artificial intelligence such as automatic driving,human-computer interaction,unmanned aerial reconnaissance and video surveillance.The target tracking algorithm based on correlation filtering has attracted a lot of researchers' attention because of its advantages of precision and speed,and has achieved remarkable results.Usually target tracking faces many challenges in the real environment: fast motion,motion blur,similar object interference,occlusion,scale change,large deformation and so on.Therefore,how to achieve fast and accurate tracking while satisfying the realtime has always been an important research topic in the field of target tracking.Aiming at the above problems,this paper proposes improved methods based on the correlation filter algorithm,including background-aware correlation filter,adaptive position correction mechanism,principal component analysis(PCA)dimension reduction and filter model update strategy.The specific work of this paper includes the following aspects.Firstly,for the fast motion and similar object interference problems in target tracking,the existing sample images are used to train more positive and negative samples through the background perception method,which reduces the tracking loss caused by the boundary effect and realizes the sample number.The increase of the sample and the improvement of the sample quality improve the classification ability of the filter in complex scenes,thereby improving the tracking accuracy of the target algorithm.Secondly,aiming at the model drift phenomenon caused by inaccurate position judgment in target tracking,a position correction mechanism is proposed,which uses high confidence condition to judge whether the position is accurate.If it is not accurate,the fusion color naming(CN)feature performs position correction.The fused features increase the ability of the algorithm to describe the target,improve the tracking accuracy of the algorithm in complex environments,and alleviate the problem of model drift.Thirdly,in order to solve the problem of increased computational complexity caused by the fusion of CN features,this paper uses PCA to reduce the dimension of the merged features.The improved algorithm comes at the cost of less tracking accuracy,which greatly reduces the computational complexity and improves the tracking speed.Fourthly,the correlation filtering algorithm is used to track the unstable phenomenon when the target is occluded and the target's large deformation.It is proposed to adaptively update the filter model by using the average peak correlation energy,multi-peak detection and maximum response value to make the algorithm better,which make the algorithm better adapt to complex scenes,and effectively track when the target is fully occluded or the amount of deformation is large.