gTRAVEL+: Enhancing Ground Segmentation for 3D Point Cloud With Temporal Noise Removal and Adaptive Plane Fitting for Urban and Off-Road Environments

Ground segmentation plays an increasingly important role as a preprocessing module for numerous applications such as Simultaneous Localization and Mapping (SLAM), place recognition, and point cloud registration. The efficiency and accuracy of ground segmentation can directly influence the performanc...

Full description

Saved in:
Bibliographic Details
Main Authors: Truong Giang Dao, Dinh Tuan Tran, Duc Manh Nguyen, Joo-Ho Lee, Anh Quang Nguyen
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Access
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10990224/
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Ground segmentation plays an increasingly important role as a preprocessing module for numerous applications such as Simultaneous Localization and Mapping (SLAM), place recognition, and point cloud registration. The efficiency and accuracy of ground segmentation can directly influence the performance of these applications. While several methods have been proposed, they continue to struggle to meet the comprehensive requirements of robust ground segmentation. To this end, we propose a ground segmentation method called gTRAVEL+, an improvement of TRAVEL that focuses exclusively on enhancing ground segmentation performance. Our method exploits a temporal and locally-aware noise removal strategy, termed Temporal Node-Wise Noise Removal (TNNR), which effectively eliminates noise points that compromise the ground plane fitting process. Additionally, Merge Node Plane Fitting (MNPF) is proposed to address issues of partial under-segmentation arising from suboptimal region sizing. Finally, Rejected Ground Node Revert (RNGR) ensures thorough geometric relationship verification, significantly improving true positives. Quantitative and qualitative experiments conducted on both urban and off-road environments show that our proposed method demonstrates superior performance, achieving a <inline-formula> <tex-math notation="LaTeX">${F} {_{{1}}}$ </tex-math></inline-formula>-score of 95.16% and competitive speed compared to state-of-the-art methods.
ISSN:2169-3536