Applying 3D Images to Object Recognition and Fetching Systems of Eye-in-hand Robotic Manipulators

• Main Authors: Shang-Hsin - Liu, 劉尚欣
• Other Authors: Sheng-Dong Xu
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/09232322285889653751

碩士 === 國立臺灣科技大學 === 自動化及控制研究所 === 105 === In this thesis, we discuss applying 3D images to object recognition and fetching systems of eye-in-hand robotic manipulators so that the robotic manipulators can automatically fetch objects. In the developed system structure, we first adopt 3D ToF (Time of flight) Depth Sensor, Kinect for Windows V2, to get 2D image data and 3D point cloud data. Then, a six DOF (Degrees-of-Freedom) industrial machinery arm, working in eye-in-hand type equipped with two-finger fixture and a Kinect at the end of the arm, is used as the hardware structure of the developing platform. In the experiments of object identification and attitude determination, we assume two special cases, i) objects are scattered, and ii) objects are stacked. In the first experiment, SURF(Speeded Up Robust Features)algorithm is used to judge if the objects are in the working area. If the answer is yes, VFH (Viewpoint Feature Histogram) algorithm is adopted to determine the attitude of the objects. Finally, the calculation results will be the basis of trajectory plan for the robot arm. In the second experiment, GC(Geometrical Consistency)and Hough Voting in CGA(Correspondence Grouping Algorithm)are used to judge if the objects are in the working area. Even though some parts of the objects located in the working area are obscures by other objects, we still can identify the target object as long as you can match enough key points. When the object is in the working area, the calculation results can be the basis of trajectory plan for the robot arm. Experiments show the feasibility of applying 3D images to object recognition and fetching systems of eye-in-hand robotic manipulators.