Apply Deep Learning and Binocular Stereo Vision For The Robotic Arm To Pick Up Objects

• Main Authors: TSAI, SHUO-LUN, 蔡碩倫
• Other Authors: TIEN, FANG-CHIN
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/zsme46

碩士 === 國立臺北科技大學 === 工業工程與管理系 === 107 === Robotic arms are an indispensable tools in automated factory applications, and many robotic arms still rely on manual compilation to complete the job. This way not only requires manual modifications but also reduces the flexibility of the robotic arm. The increase in computer computing power has enabled many machine vision methods to be applied more quickly to the industry, and the demand for machine vision has gradually shifted from 2D vision to 3D vision. In this study, the binocular stereo vision is combined with the structural light to reconstruct the three-dimensional model of the device under test, and then the robotic arm is used to complete the pick and place operation, so as to save the maintenance and the manpower of the compiler and make the robot arm more flexible. In this study, the projector firstly uses the projector to project the Gray code structure light and the phase shift structure light onto the device under test, then calculate the Gray code phase value and the four-step phase shift phase value, and compare the effects of the two structured lights on the reconstruction. Finally, select the best way to match the left and right images to reproduce the 3D model of the device under test. After obtaining the 3D information, locate the position and category of the device under test through deep learning. Lastly, convert the coordinate of the device under test in the image to the mechanical arm coordinate system by coordinate conversion to facilitate the robot to grip the device under test.