Research On Vision-based Moving Target Detection And Tracking

As an important part of the mobile robot platform to perceive the external environment,vision transmits the collected real-time images to the processing unit.After the image information is analyzed and processed,different functions are realized according to actual needs.Based on the engineering requirements,this thesis sets the human body as a moving target.The vision-based moving target detection and tracking technology detects and tracks the moving human body and applies it to the mobile robot platform to control it to follow the human body movement.The research content of this thesis is mainly divided into three modules: moving target detection module,moving target tracking module and moving target follow-up tracking control module.In order to solve the problem of detecting moving humans under dynamic background,the moving target detection module integrates two ideas of feature information detection and human body model segmentation detection combined with deep learning framework to complete the detection of human body by detecting the feature points of key parts of the human body.The detection of feature points of key parts of the human body depends on the human pose estimation algorithm.Therefore,the research in this thesis is based on the bottom-up model in the multi-person pose estimation method.First,all human key points in the image are detected by convolutional neural network for feature extraction.Then,in the prediction of the confidence map of the feature point,the method of fusion and optimization of heat map and offset is used to achieve accurate labeling of key points of the human body,and finally the human body detection results are obtained.In order to solve the problem of image tracking based on the detected moving human body,the moving target tracking module uses the relevant filter tracking method in the discriminant tracking model,the upper body of the human body in the detection result is selected as the tracking target,and the detection result is used as the input to train the relevant filters.At the same time,the real-time update enables the relevant filters to adapt to changes in the moving human body and achieve reliable tracking.In the study of related algorithms,based on the principle study of MOSSE filters,this thesis fuses the HOG features of the KCF algorithm and the scale filters of the DSST algorithm to form a fused related filter.The algorithm also solves the problem of KCF algorithm lack of scale estimation and low real-time rate of DSST algorithm,moreover,the accuracy of tracking is improved while ensuring the real-time performance of the algorithm.The moving target follow-up tracking control module uses the above tracking results as the tracking target of the mobile robot,and to realize the application of the detection module and the tracking module on the mobile robot platform.By analyzing the motion model and observation model of the mobile robot,formulate control strategy based on a monocular camera,and PID control method based on ROS system is used to control the mobile robot,eventually realized its follow-up tracking of human body motion.