A Test System of Binocular Vision of Picking Robot

Abstract

A test system of binocular vision was built for multi-degree of freedom of fruit picking positioning research in this paper. A new camera calibration method was adopted, in which the intrinsic parameters and extrinsic parameters of a camera were separated; especially the Zhang's Plane Calibration Method was used for the calibration of intrinsic parameters. In the stereo matching, firstly, according to the area feature of the object projected in these two images grabbed independently by left camera and right camera, the matching areas were determined, and then the centroid of matching areas were used as feature points to match, finally the space coordinates of the object were obtained by using triangle imaging principle. The spatial positioning tests showed that spatial positioning errors were mainly the depth positioning errors. Within the 480-640 mm, error distribution was in the range of 0-10 mm, which was the best measurement distance of binocular stereo vision. The system error of the test platform and the calibration error were the mainly influencing factors of the precision of depth recovery. With a simple segment-compensation method, the available distance of depth recovery could be extended to 390-800 mm, and most of the errors were not than 6 mm.