| Summary: | 碩士 === 國立臺北科技大學 === 自動化科技研究所 === 104 === In this paper, we propose a vision guided robot (VGR) grasping system. First, the Kinect sensor was applied to capture the 3D point cloud data. Then, the Viewpoint Feature Histogram (VFH) descriptor for 3D point cloud data encodes geometry and viewpoint, which allows simultaneous recognition of the object and registration with the stable pose on database. However, the wrong pose will be determined when the object is symmetrically placed on the viewpoint. Here, the Modified Viewpoint Feature Histogram (MVFH), is proposed to avoid the ambiguity of symmetric pose. The refine pose is further estimated with iterative closest point (ICP) after the object recognition and rough pose estimation by MVFH on database, which will decrease processing time. Therefore, the information of object pose was sent to the robot grasping system and the robot will automatically grasp the object. The experimental result shows that the proposed grasping system is efficient and affective to improve accuracy, flexibility and intelligence in VGR application.
|
|---|
