Learning Stereopsis from Geometric Synthesis for 6D Object Pose Estimation

TL;DR

This paper introduces a novel 3D geometric volume approach for monocular 6D object pose estimation, leveraging two-view geometric synthesis to improve accuracy and robustness, especially under occlusion.

Contribution

It proposes a 3D volume-based method with a coarse-to-fine framework that outperforms existing monocular approaches in 6D pose estimation.

Findings

Outperforms state-of-the-art monocular methods

Robust under occlusion and diverse scenes

Effective two-view geometric synthesis

Abstract

Current monocular-based 6D object pose estimation methods generally achieve less competitive results than RGBD-based methods, mostly due to the lack of 3D information. To make up this gap, this paper proposes a 3D geometric volume based pose estimation method with a short baseline two-view setting. By constructing a geometric volume in the 3D space, we combine the features from two adjacent images to the same 3D space. Then a network is trained to learn the distribution of the position of object keypoints in the volume, and a robust soft RANSAC solver is deployed to solve the pose in closed form. To balance accuracy and cost, we propose a coarse-to-fine framework to improve the performance in an iterative way. The experiments show that our method outperforms state-of-the-art monocular-based methods, and is robust in different objects and scenes, especially in serious occlusion situations.