Research On Target Tracking Algorithm Based On Deep Learning

With the development of computer information technology and deep learning algorithms,visual tracking technology,which is one of the popular directions in the field of computer vision,with the assistance of deep learning algorithms,has gradually become one of the important technologies in the field of artificial intelligence and the perception layer of the Internet of Things.Vision technology has become one of the important ways for humans and computers to recognize the world,and computer vision technology based on deep learning methods is increasingly applied to all aspects of military operations and social life.For example,precision-guided weapons that occupy an important position in modern warfare use visual tracking technology to improve the hit rate of targets;with the advent of the 5G era,the autonomous driving industry in the boom of artificial intelligence also uses visual tracking technology to achieve vehicle pedestrian trajectories.Tracking prediction.This article takes the visual tracking technology in computer vision technology as the background,and conducts in-depth analysis and research on the problems of single target tracking technology in visual tracking technology.The specific work is as follows:First of all,for the problem that the feature extraction network parameters of the visual tracking with siamese region proposal network algorithm are large,the algorithm structure is complicated,and the speed of the algorithm is suppressed from increasing,this paper improves and proposes a feature-based correlation based on this algorithm.A lightweight tracking algorithm for classification regression.The algorithm first adjusts the structure parameters of the feature extraction network layer and reduces the amount of network model parameters.Secondly,it performs target category and bounding-box prediction based on the feature maps of the correlation response between features.The algorithm is further simplified and performance enhanced.Secondly,in order to solve the problem that the pre-trained feature extraction network model cannot extract the target image features with good general appearance performance,this paper proposes an improved tracking algorithm that incorporates explicit context feature maps.The proposed algorithm is inspired by Bayesian inference,fuses the explicit context information of the target,and obtains a more general representative high-level semantic feature.The validity of the algorithm is verified by a high-dimensional feature dimension reduction visualization method.Third,for the visual tracking with siamese region proposal network algorithm without updating the template,the template may contain a large amount of background information,which causes the feature extraction network to extract image features mixed with a large amount of background information.A tracking algorithm that separates foreground and background features by degree,and makes full use of the potential value of foreground and background features for searching target and completes tracking tasks.Finally,based on the python development language and pytorch deep learning neural network development framework,the implementation of the algorithm is verified.Experimental results show that the algorithms proposed in this paper can achieve stable tracking of a single object.