Home // ALLSENSORS 2025, The Tenth International Conference on Advances in Sensors, Actuators, Metering and Sensing // View article

Performance Improvement of Loop Closure Detection Using Semantic Segmentation in LiDAR SLAM

Authors:
Shuya Muramatsu
Masafumi Hashimoto
Kazuhiko Takahashi

Keywords: LiDAR SLAM; NDT-Graph SLAM; loop closure detection; semantic segmentation.

Abstract:
This paper proposes a method to improve loop closure detection performance in Normal Distributions Transform (NDT) Graph Simultaneous Localization And Mapping (SLAM) using Light Detection And Ranging (LiDAR). Candidates of revisit places (loops) are detected from semantic information, such as buildings, fences, vegetation, trunks, poles, and traffic signs, based on semantic segmentation of LiDAR point cloud data. Loops are determined from loop candidates using semantic information-based point feature histograms in conjunction with conventional geometric information-based point feature histograms. Then, vehicle poses relative to loops are calculated based on the matching of point features and are applied to correct errors accumulated by NDT SLAM in the graph optimization framework. The projected-based semantic segmentation RangeNet++ is used to obtain semantic information about the surrounding environment. The experimental results obtained using the Semantic KITTI dataset demonstrate the performance of the proposed method.

Pages: 45 to 51

Copyright: Copyright (c) IARIA, 2025

Publication date: May 18, 2025

Published in: conference

ISSN: 2519-836X

ISBN: 978-1-68558-273-9

Location: Nice, France

Dates: from May 18, 2025 to May 22, 2025