Occluded object grasping based on robot stereo vision

To solve the problem of occlusion for robot grasping, a novel method is proposed, by which the robot can grasp an occluded target object accurately. In this paper, the principles of binocular vision, camera calibration, stereo rectification and stereo matching are described. By utilizing the stereo-vision algorithm, robots can obtain the three-dimensional information of objects. For accurate grasping, the depth information of the object is optimized, where the principles of connected consistency and threshold setting have been applied in the depth image. The ORB algorithm is applied to find the occlusion dynamically. The approach for grasping the occluded object is proved to be effective by experiments.