Vision-based color and shape classification using Dobot MG400: a low-cost educational robotics platform

Abstract

This study develops an eye-to-hand vision-guided robotic system using a Dobot MG400 arm, an overhead universal serial bus (USB) camera, and an Arduino-controlled pneumatic suction gripper for automated pick-and-place operations. The system detects 12 types of workpieces based on 3 shapes (circle, triangle, and square) and 4 colors (red, green, blue, and yellow) using Python and OpenCV. To ensure precision, camera-to-robot calibration is performed using a 16-point homography matrix optimized by the random sample consensus (RANSAC) algorithm. Furthermore, an adaptive correction algorithm based on affine transformation and the least squares method is implemented to compensate for coordinate deviations. Evaluated across three illumination levels (26.9, 175.4, and 430.4 lux), the correction algorithm significantly improves positioning accuracy by 48–50%, reducing residual errors to approximately 0.056–0.064 mm with a processing time of ~5.2 seconds. The system demonstrates high precision, stability, and robustness, making it highly suitable for automated small-scale industrial sorting.