Research On Workpiece Recognition And Location Technology Based On Binocular Stereo Vision

This paper focuses on identifying and locating objects by binocular stereo vision, aiming at grasping the scattered workpiece in industrial production lines and making the automated manufacturing more flexible. In this paper, hand-eye calibration, workpiece detection and identification, 3D reconstruction and workpiece grasping with collision avoidance are studied. In addition, an experimental platform to identify and grasp the workpiece based on stereo vision is built.This paper studies some calibration problems in the eye-in-hand binocular vision system. To obtain accurate localization of calibration pattern control points, a corner extraction algorithm based on perspective rectify is proposed. This algorithm can get the accurate location by undistorting and unprojecting the raw calibration pattern, resulting in higher accuracy than conventional approaches. Then the Kronecker product is employed to get the hand-eye parameter, which can eliminate the cumulative error caused by the two-step calibration process.To detect the scattered workpiece, geometric features extracted from the input image are utilized to detect objects. This paper first employs Gamma correction and bilateral filter to pre-process the image. Then an extended edge detection algorithm based on Sobel operator and region growing is proposed to obtain edges. And the geometric features are extracted by direct algebraic fitting of ellipses using gradient orientation.The support vector machine is utilized to identify the detected workpiece. In addition, a multi-feature fusion algorithm is presented to get higher identification accuracy, including HOG and LBP. To locate the workpiece, a geometric feature matching method is employed, and the matching error is also been analyzed.An experimental platform is built to accomplish the grasping task. Several experiments are conducted to verify the feasibility of this system and obtain the location accuracy, and a collision avoidance strategy is studied when the robot is grasping the object. Experimental results show that the algorithm proposed in this paper is stable, and is able to accomplish the identification and grasping task.