Object recognition and pose estimation in robotic grasping system

Currently, the warehouse automation technology is experiencing rapid growth to satisfy the increasing demand of e-commerce and provide fast, reliable delivery. Automation of the warehouse item-picking task requires the robust vision that identifies and locates objects amid cluttered environments, ma...

Full description

Saved in:
Bibliographic Details
Main Author: Zhou, Jiadong
Other Authors: Chen I-Ming
Format: Final Year Project
Language:English
Published: 2018
Subjects:
Online Access:http://hdl.handle.net/10356/75818
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:Currently, the warehouse automation technology is experiencing rapid growth to satisfy the increasing demand of e-commerce and provide fast, reliable delivery. Automation of the warehouse item-picking task requires the robust vision that identifies and locates objects amid cluttered environments, massive varieties of items and sensor noise. In this report, we present a perception system which combines deep learning and 3D shape matching techniques to overcome those difficulties. Specifically, two problems addressed in the system are: i) identification and segmentation of the objects in the scene images with a fully convolutional neural network called YOLO, and ii) determination of the objects’ 6D poses by performing geometry-based methods on the point clouds. In the end, a pick-and-place system is constructed by integrating the perception module with a motion planning module. By doing some experiments, we demonstrate that our system can reliably estimate the 6D poses of objects under a tabletop scenario.