Vision-Guided Recognition Method for Working State of Robot Arm based on Machine Learning

Abstract

To realize the capture of the target in any position within the working range of the robot arm, a vision-guided target recognition and positioning control method based on machine learning was proposed to improve the accuracy of the working state recognition. The system grabs four different contours of workpieces (triangle, pentagon, round, square and place in the designated area as the task, and by using the MATLAB to process the image information, and all the connected domains are marked bythe neighborhood areamarking algorithm. Later, the logarithmic coordinates-Fourier transform template matchingmethod is adopted to identify workpiece types and extract their centroid as the positioning reference coordinates.Combined with the 3-DOF robot arm, the standard D-H parametric method is usedto establishthe robot arm kinematics model, and through the inverse operation of the robot armand according to the position of the workpiece coordinates,the joint angle of each robot arm can be obtained. Later, it is sent to the lower microcontroller Arduino through a serial port, and then the control instruction is completed by the Arduino torealize the capture and placement of the workpiece and complete the work state recognition task.The experimental results show that the working state recognition system can meet the design requirements.