<p>To address the issue of low absolute positioning accuracy in industrial robots, this paper proposes an automatic calibration method based on stereo vision closed-loop measurement. The method aims to achieve efficient calibration and compensation of end-effector positioning errors through visual perception and kinematic optimization algorithms. By scanning a standard sphere as a point constraint and using the known distance between standard spheres as a distance constraint, an error model is established using the MD-H kinematic model, followed by a redundancy analysis of the model parameters. The Levenberg–Marquardt algorithm, incorporating redundancy parameter analysis, is employed to efficiently identify and correct the kinematic parameters of the error model. Experimental results show that this method significantly reduces the robot’s absolute positioning error, with the average positioning error decreasing by 72.12% after calibration. Compared to traditional methods, this approach demonstrates significant advantages in terms of the continuity, robustness, and computational efficiency of error compensation, making it suitable for various complex modern industrial environments.</p>