Recognition and orientation of object based on binocular vision
                                                    JIANG Meng1, WANG Yao yao1,2, CHEN Bai1,
(1.College of Mechanical & Electrical Engineering, Nanjing University of Aeronautics & Astronautics, Nanjing 210016, China;2.The State Key Laboratory of Fluid Power and Mechatronic Systems, Zhejiang University, Hangzhou 310027, China)

Abstract: Aiming at robot s lack of intelligence while working, binocular vision was researched on recognition and orientation of object. A binocular stereo vision system was built and calibrated to get parameters and relative pose of the two cameras. Descriptor based template matching was adapted, detected feature points were classified and recognized by random tree algorithm with high real time and accuracy. Depth information can be deduced with the disparity of the scene according to stereo matching between the two images. 3D reconstruction then was achieved by combining the disparity and calibrated parameters. The 2D 3D positions of the recognized feature points in the image and the real world scene were reprojected and iterated for the solution with least error, which was the desired pose of the object. The results indicate that the proposed methods can achieve accurate, fast and stable recognition and orientation of the object.

Key words: binocular vision; descriptor based template matching; stereo matching; 3D reconstruction