Research On Intelligent Control System Based On Vision For Following Service Robots

With the continuous development of artificial intelligence and robotics,more and more intelligent robots are appearing in industrial,medical,and service industries.As an important class of intelligent service robots,follow service robots have improved the quality of people’s lives and brought many conveniences to people’s daily lives.Vision-based pedestrian tracking technology for service robots is prone to target loss and tracking failure in the face of obstacle occlusion,large scale changes in targets,and interference from similar objects.In addition,uncertain parameters in the wheel-following robot system can reduce the accuracy and stability of the system’s control of the robot.In view of the above problems,this thesis researches visual target tracking and ranging technology and robot control system with indoor environment as a premise,and designs and opens a vision-based intelligent control system for service robots,as follows:Based on DCF tracking framework and Transformer model,proposed Transformer tracking prediction model based on local multi-scale features(Lo TPM),a Transformer tracking prediction model based on local multi-scale features.Design and add target position encoding,mask transformation module and local prediction modle to overcome the difficulties caused by complex scenes such as interference from similar objects,small target scale and occlusion.The Lo TPM tracker is experimented on multiple SOT datasets with real cameras.The experiments demonstrate that the Lo TPM tracker can accurately track the target pedestrians in complex working conditions while satisfying the tracking real-time performance.Complete the development and experiments of Orbbec Astra Pro depth camera ranging algorithm.Study the principle of distance measurement of the depth camera,conduct calibration experiments on the depth camera,obtain the camera parameters,and use the structured light ranging algorithm to complete the distance measurement of target pedestrians.The experiments prove that the measurement accuracy of Orbbec Astra Pro depth camera meets the following robot ranging requirements and lays the foundation for the vision-based following robot system.The sliding mode control of variable speed reaching law method based on stochastic configuration networks is proposed for following robot control.According to the characteristics of the system,the following robot dynamics model is established,and the controller is designed by using backstepping control and sliding mode control.The switching function dsat proposed to reduce the chattering of the system.The fast approximation capability of stochastic configuration networks for stochastic parameters is utilized to observe the uncertainty terms and compensate for the errors in the system.Simulation experiments of the established control method using MATLAB and Simulink are conducted to demonstrate that the proposed control method can effectively observe the uncertainty terms in the system,accurately compensate the errors,and improved the control performance of the following robot.The hardware experimental platform of following robot is built,and the software for the visual tracking system of following robot is developed.The target tracking system of the following robot is developed under ROS system with indoor scenes as the premise,and the program is written.Finally,a variety of experimental scenarios are designed for experiments to prove that the robot can stably track the target pedestrian in a variety of scenarios.