Research On Visual Tracking Algorithm Based On Correlation Filter

In recent years,correlation filter based tracking algorithms have gained much attention and research of the majority of scholars,due to their superior accuracy and speed,and have also achieved remarkable research results.However,it still undergoes many interferences and challenges of many external factors,such as occlusion,illumination variances,scale changes and fast motion,which will directly affect the performance of the tracking algorithm.In order to improve the accuracy and success rate of tracking algorithm,this paper mainly analyzes and investigates the correlation filter algorithms,and makes full use of its performance advantages to realize a series of innovations and improvements.The main research contents and innovations are given as follows:(1)Since the traditional single feature target tracking algorithm cannot adapt to the change of complex scene well.Based on the context-aware correlation filter,a multi-scale correlation filter tracking algorithm based on adaptive feature fusion is proposed.The algorithm fully considers the characteristics of multiple feature performance complementarities,and proposes an adaptive feature fusion method to achieve accurate translation prediction.In order to improve the scale adaptability of the algorithm,a scale discriminative correlation filter is introduced to achieve accurate scale estimation.In addition,in order to improve the update quality of the tracking model,an adaptive model updating method is proposed,which is performed by presetting the response threshold as the judgment condition of the scale filter and translation filter update model,the model quality will be improved and the tracking drift problem will be relieved to some extent.Extensive experimental results show that the proposed algorithm achieves superior performance in both accuracy and success rate,and has better robustness in sophisticated scenarios such as scale change,deformation,fast motion and occlusion.(2)In order to address the problem of external factors such as occlusion,out-of-view and scale change encountered in the long-term tracking process,a long?term correlation tracking via spatial–temporal context is proposed.We train two discriminative correlation filters for achieving long-term object tracking,which is achieved by learning a context-aware filter for translation estimation,and achieved by learning a scale correlation discriminative correlation filter to estimate the scale change from the best confidence results.In addition,we propose an effective model update strategy to alleviate the unrecoverable drift caused by noise update,and propose an efficient redetecting activate strategy to improve the robustness under long-term tracking failure.Extensive experimental results show that the proposed algorithm performs favorably against the state-of-the-art tracking methods.(3)In order to improve the adaptability of tracking algorithm in some sophisticated appearance changes,and enhance the discriminative performance of classifiers,a hyper-feature fusion visual tracking algorithm based on spatio-temporal context is proposed,which combines traditional hand-crafted features and the deep features,and an adaptive feature fusion method is proposed to achieve accurate translation estimation.The output constraint transfer optimization method is introduced to control the correlation output response map to follow the Gaussian distribution,which gain the the robustness to target appearance variations.In addition,we propose an effective model update and scale variation strategy to mitigate the tracking drift caused by model pollution,which improve the robustness of the model in sophisticated scenarios such as fast motion,motion blur,occlusion,out-of-view and large-scale changes.Extensive experimental results show that the proposed algorithm performs more robust than the state-of-the-art methods,and has achieved superior success rate,accuracy and robustness.