Research On Target Tracking Algorithm Based On Kernel Correlation Filtering

Object tracking is one of the most challenging tasks in the field of computer vision.So far,researchers have proposed a number of excellent target tracking algorithms,but the target tracking in the actual environment is still facing a series of challenges.The change in the process of target tracking changes in appearance and the surrounding environment often leads to the failure of tracking,including scale change,illumination variation,rotation,all or part of occlusion.In order to solve the above problems,this paper aimed at nuclear related filter tracking algorithm is studied,and the combination algorithm of convolutional neural network to improve the nuclear correlation filtering algorithm.The main work is as follows:1)Proposed a correlation filter tracking algorithm for advancedKCF heavy,is used to solve the robustness problem of kernel filter tracking algorithm under occlusion.First proposed a target detection mechanism with occlusion and lost,can accurately determine the current frame tracking effect;and then presents an inverted Pyramid candidate the search box,according to the region is given different search step,greatly improves the search speed of the candidate frame;finally,an adaptive template update method,the greatest degree of assurance within the template contains is the effective information of the target,which can make the algorithm still has very strong adaptability in the occluded environment2)Proposed a nuclear related filtering fusion convolutional neural network KCF_GOTURN tracking algorithm,the off-line training and on-line updating organically,can from the mass database to obtain useful information for target tracking,real-time state and target can be obtained.The sidelobe ratio for the bridge,a KCF algorithm is proposed and GOTURN the fusion algorithm,the KCF algorithm when the poor state of tracking,GOTURN algorithm is used to correct it;at the same time,put forward the way of updating an adaptive KCF algorithm and GOTURN algorithm template,to ensure maximum containing KCF template are effective information of the target,but also to ensure that the network input is GOTURN the correct target tracking,the algorithm has strong robustness to various complex environments.3)According to the above algorithm,this paper formed the final algorithm.The improved algorithm is tested using all 50 video sequences standard test set,the video sequence contains the object occlusion,scale change and rotation of different interference factors.The experimental results show that the precision and accuracy of the KCF_GOTURN algorithm of overlapping distance were 0.716 and 0.570,compared with advancedKCF algorithm,were increased by 0.049 and 0.032;compared with the original KCF algorithm;were increased by 0.073 and 0.047,compared with the original GOTURN algorithm,were increased by 0.149 and 0.132.The experimental results show that this algorithm is better than the current mainstream of the target tracking algorithm can adapt to the changes in the appearance of the target in the complex environment,still can track the target is stable and accurate.