Research On Video Target Tracking Algorithm Based On Correlation Filtering

As an essential research subject in the field of computer vision,the core task of target tracking is to estimate the position and scale of the target in the subsequent video sequence after determining the position and scale of the target in the initial frame,so as to realize the tracking of the target trajectory.The correlation filtering algorithm takes the ridge regression model as the basic framework,makes full use of the circulant matrix to sample samples and converts complex operations in the time domain to the frequency domain for solution,and has excellent performance in terms of accuracy and real-time performance.However,in actual tracking scenes,there are many unpredictable tracking challenges,such as fast motion,severe deformation,occlusion and other interference factors.Therefore,this article mainly focuses on the improvement of the tracking model,and effectively combines the regularization technique in machine learning,and verifies the validity of the proposed algorithm through a great deal of experiments.The content of this paper mainly includes the following aspects:Firstly,aiming at the problem that many correlation filtering algorithms study all feature channels with equal weight,ignoring that some channels have a large amount of redundant information,resulting in the algorithm can not effectively use features to reduce the tracking and positioning accuracy,a spatio-temporal regularized target tracking algorithm based on channel reliability is proposed.By constructing the channel regularization term,the weights of different feature channels are solved in the training stage to realize the weighting of different feature channels,so as to further improve the positioning accuracy.Secondly,aiming at the problem that the spatio-temporal regularized target tracking algorithm expands the search area in order to alleviate the boundary effect,the filter is more inclined to learn from the background,which leads to the target is easy to fail to track when experiencing the challenges of occlusion,similar background interference and so on,a spatio-temporal regularized target tracking algorithm based on aberrance repression is proposed.By constructing the aberrance repression regularized term,we can repress the aberrances in the detection process and prevent the accumulation of errors caused by tracking drift.Experimental results show that the proposed algorithm is superior to other comparison algorithms in terms of tracking accuracy and robustness.Finally,aiming at the problems that the spatial regularied correlation filtering algorithm cannot accurately model the appearance of the target using a single feature,the spatial regularied weight is fixed,and the model is easy to degenerate,a target tracking algorithm based on saliency awareness and consistency constraint is proposed.Multi-feature fusion is carried out between manual features and deep features extracted from lightweight network to meet the real-time requirements on the premise of improving the tracking accuracy.The saliency detection algorithm is used to obtain the saliency detection results of the initial frame,and the change information of the target is introduced into the regularized weight graph to construct the significant perception adaptive spatial regularization term to dynamically update the regularized weight sequence,which can alleviate the boundary effect more effectively.Minimize the difference between the actual consistency response and the ideal consistency response,and dynamically adjust the strength of the consistency constraint by judging the quality of the response map to effectively prevent the filter template from degrading.The proposed algorithm and other excellent algorithms are tested on three different datasets:OTB2015,Temple Color128,and UAV20 L.The experimental results show that the algorithm has achieved more ideal results in terms of accuracy and success rate.