Research On Online Detection Of Movable Objects And Dynamic Gripping By Robot Arm Based On Machine Vision

Compared with traditional manipulators,collaborative manipulators have made comprehensive improvements in terms of movement perception,path planning,and grasping efficiency,and they are being more and more widely used in various industrial production lines.At present,the main constraint factor of collaborative robotic arm grasping technology is that the level of intelligence needs to be improved,which is also the research focus of researchers in recent years.This paper introduces machine vision as the "eye" of the robotic arm to conduct research on the dynamic grasping technology of the robotic arm,and introduces the sorting of fruit models on the conveyor belt as a research case,which is mainly used in the field of robotic arm grasping teaching.Firstly,the hardware of the visual system and robotic arm system is selected and designed.The depth camera is calibrated to obtain internal and external parameters and eliminate lens distortion.The Aruco code is used for hand eye calibration to complete the conversion from the camera coordinate system to the robotic arm base coordinate system,so that the calibration error is within 1cm,preparing for subsequent grasping.Secondly,traditional object detection algorithms have single features and are suitable for static scenes,resulting in a decrease in detection accuracy for dynamic scenes.This paper proposes a deep learning object detection model based on YOLOv5,which uses CSPDarket53 as a feature extraction network and combines it with the Head layer to output prediction results.The experiment shows that the m AP value of YOLOv5 prediction results is 95%,which can accurately predict the types of fruit models in the image.Then,TOF time flight technology is used to obtain a depth map within the camera’s field of view,align the depth map with the color map,and obtain the target coordinates based on the measured depth values and the previously calibrated results.Third,design a real-time monitoring algorithm for moving targets,use the encoder to monitor the instantaneous speed of the conveyor belt,and obtain the position of the moving target in real time based on time and speed;design a strategy for manipulator control and dynamic tracking path planning based on the S-type acceleration and deceleration algorithm,effectively reducing the number of manipulators.Vibration during joint movement,door-shaped motion trajectory planning is used in the process of placing the target,which effectively avoids obstacles while ensuring stability.Finally,the hardware and software platforms for the visual system and ROS robotic arm grasping system were built,and a human-machine interface was constructed for data interaction.Finally,experimental verification showed that the average positioning accuracy of the moving target was at the millimeter level,and the grasping efficiency of the robotic arm was above 92%.It was able to stably sort fruit models,meeting the design requirements of the project.