Pose Estimation by Removing Negative Surface on the Next Viewpoint

View at publisher Summarise with VIRAT
Authors: Tomoki Mizuno, Tsuyoshi Tasaki
Cited by: 0
References: 22

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

We have developed a new pose estimation neural network that focuses on surfaces that become newly visible with the viewpoint movement. Predicting the next viewpoint for pose estimation and moving there improves pose estimation accuracy. However, when there are a few newly visible surfaces after viewpoint movement, the accuracy improvement is limited. To address this, we have developed a novel method making masked images removing negative impact surfaces that was visible before viewpoint movement, allowing the neural network to focus on the newly visible surfaces. As a result, pose estimation accuracy improved. Experimental results showed that using the masked image improved the success rate of pose estimation by 3.4 points compared to the conventional method without the masked image. Furthermore, our method enables the robot to display 89.5% of products, which is an improvement of 5.3 points over the previous method.

Fields of Study & Subjects

Robot Manipulation and Learning 3D Shape Modeling and Analysis Advanced Neural Network Applications

Publication details

Journal / source
IEEJ Transactions on Electronics Information and Systems
Volume
146
Issue
4
Pages
353–359
Publication date
31 March 2026
Language
EN