Design And Implementation Of A Vision-based Cleaning Robot Pose Detection System

Visual pose detection is an emerging technology based on machine vision.In order to obtain the target feature information in real time by the end effector of the arm,the precise spatial position of the target relative to the arm is detected,thereby driving the arm to move.In this paper,a machine vision-based cleaning robot arm attitude detection system is designed and implemented.The system mainly realizes the relative spatial position detection between the robot arm and the solar panel and the calculation of the end position of the arm.Firstly,through the kinematics analysis and modeling of industrial manipulators,the forward solution of the kinematics of the manipulator is obtained by the coordinate transformation method and the inverse solution is obtained by the inverse transformation method.Secondly,the visual pose transformation relationship of the robot arm is established,the camera internal reference calibration is performed,the pose coordinate conversion in the imaging and imaging process of the monocular camera is analyzed,and the image pre-processing,edge detection and connected domain are performed on the collected solar panel image.Acquire image features such as detection.Finally,the camera calibration parameters and image feature information are brought into the perspective geometric relationship of the spatial points to calculate the perspective projection relationship between the target object and the image point in the system.The target end pose detection method is used to calculate the end effector of the robot arm.The proximity vector relative to the solar panel further determines the target pose of the robot arm.The multi-degree-of-freedom solar panel cleaning robot arm control system is usedto perform visual pose detection experiments,and the real-time acquired solar panel feature point information is introduced into the robot arm motion control,thereby completing the pose detection of the target board and the robot arm.The experimental results show that the cleaning robotic visual attitude detection system can detect the relative spatial position between the solar target plate and the robot arm in real time.When the detection distance is within 50-800 mm,the position error accuracy can be up to 2.7%.