Study On Visual Perception And Following System For Pedestrian Target Of Wheeled Mobile Robot Based On ROS

In recent years,mobile robot technology has developed rapidly.Especially with the development of artificial intelligence technology,mobile robot visual perception technology has become a hotspot,and it has a wide application prospect in many fields.Visual perception algorithm based on target detection,tracking and ranging has the advantages of accuracy,efficiency and intelligence.Applying it to the application of mobile robots will greatly improve the intelligent perception ability of mobile robots and make accurate and fast decision-making control for mobile robots.In this paper,the wheeled mobile robot applied in indoor environment is studied,and a visual perception and tracking system of pedestrian target based on ROS is constructed.The perception algorithms of pedestrian target detection,tracking and ranging are mainly studied.The transplantation optimization is carried out on the mobile robot system.After that,the accuracy and rapidity of the perception algorithm and the feasibility of implementation on the mobile machine are tested by experiments.Finally the mobile robot completed the task of fast and accurate visual perception and simple tracking of the surrounding pedestrian target.The main contents of this paper are as follows:1)The hardware and software system of the mobile robot is constructed.Firstly,the parallel structure four-wheel chassis based on McRam wheel is designed according to the flexible mobile requirements of the mobile robot.Then,the hardware system of the mobile robot is constructed,including the underlying control hardware system based on STM32 and the upper perception hardware system based on TX2 embedded platform.Finally,the ROS software system is constructed,including ROS-based system software framework and functional implementation process.Through the construction of hardware and software system,it lays a foundation for the realization of the perception and tracking algorithm of the mobile robot.2)The pedestrian detection and tracking perception algorithm based on convolutional neural network is studied and tested.Firstly,the pedestrian detection and perception algorithm based on YOLOv3 framework is studied.Aiming at the problems of jamming,slow detection speed and large memory occupied by network model after transplanting YOLOv3 algorithm to TX2,the pedestrian data set is added to retrain the YOLOv3 network and the TensorRT is used to optimize the reasoning acceleration of the YOLOv3 training model,for improving the pedestrian detection accuracy and realtime performance.YOLOv3 network is retrained and TensorRT is used to optimize the model of YOLOv3 training.The performance of the optimized pedestrian target detection algorithm is greatly improved compared with that before optimization.Then DeepSort pedestrian tracking algorithm is studied and combined with YOLOv3 to achieve fast and accurate pedestrian target tracking.Finally,the above algorithm is tested and tested on TX2.The accuracy and real-time performance of the optimized pedestrian detection and tracking perception algorithm are tested.Through the research and implementation of the above perception algorithm,it provides an algorithm basis for the program implementation on the ROS software platform of the mobile robot.3)The pedestrian ranging perception algorithm based on binocular camera is studied and tested.In order to obtain more accurate ranging accuracy,the binocular camera is firstly calibrated according to Zhang Zhengyou's calibration algorithm and the calibration results are obtained.Then the matching experiment based on SGBM binocular matching algorithm is carried out,and a better matching effect is obtained.Finally,the ranging accuracy of this ranging algorithm on the mobile robot is tested by human target ranging experiment.The experimental results show that the ranging algorithm has high accuracy and can meet the ranging perception requirements of the mobile robot,and lays the foundation for its transplantation on the mobile robot.4)The realization of the perception and tracking system is studied and the whole function experiment is carried out.Firstly,different perception and following motion control strategies are formulated according to whether there are pedestrian targets around the mobile robot and whether there are single or multiple targets.Then,the underlying control system is programmed according to the following control requirements,and the ROS-based upper layer perception software system program is implemented according to the perception requirements.Finally,based on the mobile robot sensing and following system,the overall function is tested and verified,which verifies the validity of the perception algorithm and the feasibility of the tracking motion control strategy of the mobile robot.Through the realization of the above functions,the goal of visual perception and tracking motion control of the mobile robot for pedestrian targets is achieved.