Target Position and Posture Recognition Based on RGB-D Images for Autonomous Grasping Robot Arm Manipulation

Abstract

Target position and posture recognition play a key role for the autonomous grasping manipulation of the robot arm. This paper proposes a novel target recognition method based on the RGB-D image. The RGB-D image is achieved by combining the target RGB image with its depth image. The combination is realized by means of the joint coordinate calibration. In order to recognize the target, this paper firstly establishes the target template by placing the target object on a reference position in the experimental scene. Furthermore, an improved Iterative Closest Point (ICP) algorithm is employed. By matching the real-time target RGB-D image with the established target template, the target position and posture can be recognized accurately. The proposed method has been applied in an experimental robot arm to grasp the object autonomously, and the result shows that the proposed method is effective.