| The research of object tracking algorithm not only promotes the development of computer vision theory,but also is widely used in life,which has important application value.Due to the complexity of challenges encountered in the tracking process,which requires robustness of the algorithm to be improved.Based on the popular target tracking method and the successful application of deep learning in the field of image processing,this paper proposes a target tracking algorithm that combines the deep learning method with the convolution filter framework.Firstly,aiming at the problem that a single feature can not well model the appearance of the target,the feature fusion and the convolution network of deep learning model can improve.The deep learning method can be used to obtain the features of spatial information and semantic information.Then,the output of the convolutional layer is used to train the correlation filters and results of filters are weighted fused to find the location of the maximum response.Secondly,considering that the framework uses a fixed-size search window,the relative movement between the target and the camera during the tracking process will cause scale changes.Therefore,this paper introduces the method of scale estimation.By learning multiple candidate samples,adaptive tracking can be realized.Finally,in order to improve the tracking accuracy of the algorithm,the confidence level of the detection is compared with the threshold.If it is greater than the threshold,it is taken as the target position.This will help improve the accuracy of the positive and negative samples.In this paper,the improved algorithm is tested on a large number of video sequences.At the same time,this paper also uses camera to track the target in real time and analyzes the algorithm in different situations.Experimental results show that the tracker has good performance in occlusion,fast motion and illumination conditions. |
