Please use this identifier to cite or link to this item:
|Title:||Object recognition and pose estimation for bin-picking||Authors:||Zhou, Lingjin||Keywords:||DRNTU::Engineering::Aeronautical engineering||Issue Date:||2019||Abstract:||3D object localization and pose estimation have been studied extensively in bin-picking problems. To solve these problems, the vision system plays an important role. In this report, the author proposes an approach of vision system that is implemented on the Robot Operating System (ROS) platform. It incorporates GMS Feature Correspondence Method, K-Means clustering and Line Segment Detector (LSD) to identify the objects inside a bin and obtain the pose estimation of the objects. In this approach, 2D images are processed with Open Source Computer Vision Library (OpenCV) and the 3D point clouds are processed with Point Cloud Library (PCL). The aim of this approach is to detect the objects and find the 6-Degree of Freedom (6-DoF) pose. The estimated pose information is essential for motion planning and robot execution. This approach is designed for random bin-picking of multiple identical circular objects and is robust in complex environments. The experiments show that this approach is efficient and fast and is applicable in real-time bin-picking task inside a warehouse. Different from the current solutions, this approach does not require any CAD data or 3D modelling of the objects. In this approach, only a dataset of 2D images of the objects is necessary for object detection. Therefore, this approach provides a simple and efficient solution to the vision system of the bin picking problem. The testing results prove the reliability and robustness of this approach.||URI:||http://hdl.handle.net/10356/78549||Rights:||Nanyang Technological University||Fulltext Permission:||restricted||Fulltext Availability:||With Fulltext|
|Appears in Collections:||MAE Student Reports (FYP/IA/PA/PI)|
Items in DR-NTU are protected by copyright, with all rights reserved, unless otherwise indicated.