基於機器學習之語義分割及幾何建模配對進行物件姿態估測及夾取

• Main Authors: SUN, GUO-JHEN, 孫國禎
• Other Authors: LIN, HUEI-YUNG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/kr9qd2

碩士 === 國立中正大學 === 電機工程研究所 === 107 === In the automatic control, the use of computer vision technology to detect and identify objects and then operate them by robotic arms is a main core technology. In robotic manipulation, one of the fields is visual servo control. The combination of computer vision and automatic control will be the main core technology of automation factories or fully automatic stores. In this thesis, we propose a two-dimensional and three-dimensional information combined with computer vision through a robotic arm. Based on two-dimensional information, we use machine learning to identify specific objects and obtain pixel-level segmentation mask results. We use the basic geometric model to estimate the pose of the object instead of preconstructing the accurate CAD model to select grasping points. By using computer vision data to control the robot arm, we use the eye-in-hand architecture to obtained visual data from a camera mounted directly on the robot arm, enabling the arm to be positioned in a real environment to complete the whole robotic gripping system. In the experiment, the object segmentation results will be displayed and use the basic geometric object matching to find the object grasping points, and perform hand-eye calibration to find the actual object position to perform the grasp work.