Research On Target Tracking Method Based On Correlation Filter And Depth Feature

Visual target tracking is an important topic in the field of computer vision research.At present,researchers in the field of visual target tracking have proposed a series of excellent algorithms and frameworks,but in view of the deformation,occlusion,illumination changes,random motion and other issues in the target tracking process.To ensure the real-time,accuracy and robustness of the tracking algorithm still faces great challenges.Aiming at the above problems,this paper proposes an improved target tracking algorithm based on the classical target tracking framework.Firstly,we analyze the different roles of the target features of the different layers of the convolutional neural network in the tracking task,and then study the possibility of using the more accurate convolutional neural network model to improve the accuracy of the target tracking results.On this basis,this paper analyzes the Res Net-101 deep convolutional neural network model.Because the output feature mapping of different network layers has different characteristics,this paper finally selects the output feature of three network layers as the target feature to detect and track the target more accurately,and to cope with the possible challenges of target deformation,shielding and illumination.the neural network training process usually only initializes the network weight once.This paper compares the effects of different weight initialization methods on network accuracy,and proposes a progressive weight initialization method to improve network performance.At the same time,the sampling of the target image and the updating of the tracking model parameters in the existing target tracking algorithm are all updated frame by frame.By comparing and analyzing the image characteristics of adjacent frames,this paper improves the sparse method to sample the sample image and update the model.Finally,this paper builds an experimental environment for target tracking,tests the performance of the proposed tracking algorithm on the VOT target tracking standard data set,and compares the performance with that of other mainstream algorithms.The results show that the improved method proposed in this paper can effectively improve the tracking performance.