Title:
RPN-based architecture for object detection and pose estimation using RGB-D data
RPN-based architecture for object detection and pose estimation using RGB-D data
Author(s)
Gourdon, Remi
Advisor(s)
Collins, Thomas R.
Editor(s)
Collections
Supplementary to
Permanent Link
Abstract
Pose estimation is a topic of important research in the fields of robotic and computer vision, particularly for applications in autonomous transportation and robotic manipulators. This thesis presents the implementation of a pose estimation network capable of leveraging color and depth information from commercial off-the-shelf sensors, and proposes its integration as an extension to well-known architectures based on Region Proposal Networks. This work also presents an automated image and pose data collection method using an industrial robotic arm and multiple cameras, and describes its use for the acquisition of a chicken dataset as part of a research effort in poultry processing automation. The estimation results obtained on this application-specific dataset are presented.
Sponsor
Date Issued
2018-12-07
Extent
Resource Type
Text
Resource Subtype
Thesis