Research And Implementation Of Robot Vision Tracking Algorithms Based On Deep Learning

Robot target tracking technology is more and more widely used in robot navigation and positioning,human-computer interaction,unmanned driving and other fields.Among them,visionbased service robots also play an important role in life.This paper designs and implements an intelligent robot target tracking method based on deep learning by using monocular camera in the application scenario of museum explanation and shopping guide.The most important point in the target tracking model is the feature extraction of the target.This paper selects some deep convolutional neural network algorithms with excellent current feature description ability,and conducts in-depth comparative research.In this paper,the existing GOTURN and TLD monocular vision tracking algorithms are tested on the robot.It is found that GOTURN algorithm is lack of tracking failure detection mechanism.At the same time,the traditional TLD algorithm has the advantages of being susceptible to illumination during the tracking process.In order to ensure the tracking accuracy and real-time performance,an improved GOTURN-LD algorithm is proposed in this paper,which uses the offline training depth learning model for online tracking and integrates it with the traditional TLD algorithm framework with detector.In addition,the detection module of GOTURN-LD algorithm is improved,and the area recommendation algorithm is added to reduce the scanning area,so as to reduce the computation of detection,speed up target occlusion or recovery after loss and further improve the tracking accuracy of the algorithm.In this paper,the improved tracking algorithm is applied to the actual scene experiment of the robot system,which mainly completes functions such as robot motion,information collection and processing.By setting up the test environment and testing the target under different illumination and occlusion,the tracking performance of the improved algorithm is verified.