Target Tracking Research Of Adaptive Feature Fusion

Target tracking is a key problem in computer vision.The tracking accuracy of traditional correlation filtering algorithm is poor when the target is occluded and the scale changes rapidly,while the tracking algorithm of correlation filtering combined with deep learning has a long training time,so it is unable to track the target in real time.Aiming at the above problems,A target tracking research of adaptive feature fusion is proposed.Firstly,the two basic features of the target are extracted and fused with different weights,so as to obtain the initial fusion features with different expression preferences.The reliability of the initial fusion features was judged,and the fusion features with high reliability were selected for the second adaptive fusion.The fusion weight was determined according to the reliability value of the initial fusion features.In the second fusion feature,the credibility strategy is used to select the optimal feature as the tracking feature of the current frame target,and estimate the candidate position of the target.Secondly,occlusion judgment is made on the response degree of the candidate position,if the result is higher than the preset re detection threshold,it indicates that the estimated target position is less reliable,and the occlusion re-detection mechanism is started to re detect the target position.On the contrary,the candidate position is highly reliable,which is the target position and output as the tracking result.Finally,aiming at the problem that the error accumulation will lead to the poor tracking effect in the long-term tracking process,the preset update threshold can update the model adaptively to improve the accuracy of the target description of the model.In order to verify the tracking effect of the algorithm,the video sequence in the international standard database is compared with the current mainstream algorithm.The experimental results show that the two feature fusion can improve the description ability of the feature to the target,reduce the occurrence of tracking drift,and the occlusion recheck mechanism can effectively judge the occurrence of occlusion of the target,so as to accurately track the local occlusion of the target.In addition,the algorithm not only improves the tracking accuracy of the correlation filter tracking algorithm,but also does not significantly reduce the tracking speed,ensuring the real-time tracking of the target.There are 23 figures,8 tables and 66 references in this paper.