A Symmetric Keypoint Generation Framework for Dense Surface-Aware 6DoF Pose Estimation
Loading...
Authors
Jadhav, Akash Avinash
Date
2025-08-27
Type
thesis
Language
eng
Keyword
Computer Vision , Deep Learning , 6DoF Pose Estimation
Alternative Title
Abstract
We present a novel method for 6DoF object pose estimation that combines dense surface point prediction with symmetry-aware keypoint learning. Our approach predicts per-pixel radial distances from RGB-D images to a minimal set of four 3D keypoints, and uses a Direct Linear Transform (DLT) formulation to estimate accurate object-centric surface coordinates. To address the challenge of symmetric objects, where fixed keypoint orderings often lead to inconsistent regression, we first introduce a manual symmetric keypoint generation method based on Oriented Bounding Boxes (OBB), which ensures spatially balanced and viewpoint-consistent keypoints across symmetric poses. Building upon this, we propose a Dense Graph Convolutional Network (DGCN) that learns symmetric keypoints directly from the object mesh, generalizing the OBB strategy by optimizing for symmetry consistency and geometric stability. Combined with a per-pixel radial regression network and RANSAC-based refinement, our framework produces accurate and robust pose estimates even in cluttered or occluded scenes. Evaluations on LINEMOD, Occlusion LINEMOD, and YCB-Video show that our method achieves state-of-the-art performance, with substantial improvements on symmetric object categories.
Description
Citation
Publisher
License
Queen's University's Thesis/Dissertation Non-Exclusive License for Deposit to QSpace and Library and Archives Canada
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.
Intellectual Property Guidelines at Queen's University
Copying and Preserving Your Thesis
This publication is made available by the authority of the copyright owner solely for the purpose of private study and research and may not be copied or reproduced except as permitted by the copyright laws without written authority from the copyright owner.