Research On Target Tracking Algorithm Based On Deep Learning

Target tracking is an important research topic of computer vision,which is widely used in security monitoring,human-computer interaction,smart city and Internet of Things.The advent of the era of big data and the rise of deep learning methods provide a new opportunity for target tracking research.The target tracking algorithm based on deep learning with stronger ability of extracting target features has higher accuracy and robustness in different environments.In this paper,in order to balance the accuracy and real-time performance,the SiamFC algorithm is selected as the infrastructure,tracking network architecture and tracking strategy of SiamFC algorithm are improved.The main works are as follows:In terms of network architecture,aiming at soloving the problem that the SiamFC algorithm can't extract robust features with a shallow network,target tracking algorithm with a twin deepened neural network based on multi-layer feature fusion is proposed.By means of transforming the network architecture from a single branch structure to a branch structure,the algorithm realizes the organic fusion of spatial features and semantic features,so that the convolution features contain more information,and the target can be tracked more accurately.In terms of tracking strategy,for the problem of poor tracking effect of SiamFC algorithm in complex color scenes,target tracking algorithm with a twin deepened neural network and a matching mechanism is proposed.Color histogram matching mechanism is introduced to calculate the matching rate between candidate samples and template.The candidate sample with higher matching degree is selected as the tracking result,which improves the algorithm's discriminating ability.A comparative test was conducted on the public data set OTB2015.The test results show that the algorithms proposed in this paper not only guarantee the real-time tracking of the target,but also improve the accuracy and robustness.