GSeg3D: A High-Precision Grid-Based Algorithm for Safety-Critical Ground Segmentation in LiDAR Point Clouds

Muhammad Haider Khan Lodhi; Christoph Hertzberg

cs.ROMar 4, 2026

GSeg3D: A High-Precision Grid-Based Algorithm for Safety-Critical Ground Segmentation in LiDAR Point Clouds

Authors: Muhammad Haider Khan Lodhi, Christoph Hertzberg

Organizations: German Research Center for Artificial Intelligence (DFKI), Robotics Innovation Center, Bremen, Germany

Abstract

Ground segmentation in point cloud data is the process of separating ground points from non-ground points. This task is fundamental for perception in autonomous driving and robotics, where safety and reliable operation depend on the precise detection of obstacles and navigable surfaces. Existing methods often fall short of the high precision required in safety-critical environments, leading to false detections that can compromise decision-making. In this work, we present a ground segmentation approach designed to deliver consistently high precision, supporting the stringent requirements of autonomous vehicles and robotic systems operating in real-world, safety-critical scenarios.

Figures & tables

Fig. 1 : GSeg3D: The point cloud is segmented into ground points (magenta) and non-ground points, which are color-coded according to their height to visualize vertical structures.

Fig. 2 : Each phase consists of a four-step process. The figure displays the raw point cloud (in white), the detected ground points (highlighted in magenta), and the non-ground points, which are color-coded according to their height along the positive z-axis.

Fig. 3 : Phase I uses grid cells with large height to capture tall structures as non-ground points. The coarse ground points from Phase I are subsequently processed in Phase II using grid cells with small height.

Fig. 4 : Correction of over-segmentation of ground points based on dual-phase segmentation.

Step	First Phase	Second Phase
Grid Representation	Large cell height	Small cell height
Local Eigen Classification	Identical	Identical
Surface Gradient Analysis	Identical	Identical
Ground Region Expansion	Neighbor expansion	Neighbor expansion and additional point-wise check

TABLE I : Processing steps in 1 st and 2 nd segmentation phase.

Fig. 5 : Points (black) assigned to a grid cell with dominant eigenvectors (green) are shown. The cell is classified as Line (a), Planar (b), or Non-Planar (c) based on the local eigen classification.

Fig. 6 : To estimate the local surface slope, a plane model is fitted within each grid cell, allowing for a detailed and localized analysis of the terrain’s inclination.

Method	00			01			02			03
Method	Pr.	Rc.	F1	Pr.	Rc.	F1	Pr.	Rc.	F1	Pr.	Rc.	F1
CGSeg	91.6 $\pm$ 9.1	71.0 $\pm$ 10.6	80.0	97.2 $\pm$ 2.9	75.7 $\pm$ 10.5	85.1	91.8 $\pm$ 7.7	70.0 $\pm$ 7.7	79.4	92.4 $\pm$ 8.0	69.0 $\pm$ 8.1	79.0
GPF	95.1 $\pm$ 4.9	80.9 $\pm$ 25.5	87.4	93.9 $\pm$ 6.4	42.6 $\pm$ 23.5	55.7	96.4 $\pm$ 3.5	79.0 $\pm$ 18.5	86.8	97.5 $\pm$ 2.8	78.7 $\pm$ 13.5	87.0
Linefit	98.2 $\pm$ 1.0	81.1 $\pm$ 10.0	88.8	99.0 $\pm$ 1.4	74.6 $\pm$ 9.6	84.9	98.0 $\pm$ 1.2	82.5 $\pm$ 5.9	89.5	95.8 $\pm$ 4.1	79.3 $\pm$ 8.4	86.6
Patchwork	92.2 $\pm$ 3.5	93.3 $\pm$ 3.9	92.7	95.6 $\pm$ 3.5	89.8 $\pm$ 5.2	92.6	92.1 $\pm$ 3.9	93.2 $\pm$ 3.8	92.6	89.4 $\pm$ 7.1	95.3 $\pm$ 2.7	92.2
Patchwork++	93.8 $\pm$ 3.9	93.9 $\pm$ 3.3	93.9	89.9 $\pm$ 6.4	97.4 $\pm$ 2.8	93.5	93.8 $\pm$ 3.9	93.6 $\pm$ 3.9	93.7	94.7 $\pm$ 2.9	91.0 $\pm$ 6.6	92.8

TABLE II : Comparison of ground segmentation methods on SemanticKITTI (sequences 00–10). Results are reported as mean

\pm

std (%).

Method	Pr.	Rc.	F1	Rt.	Pr. Std	Rc. Std	Rt. Std
CGSeg	92.1	70.3	79.7	59.1	8.8	9.8	11.0
GPF	95.3	75.1	83.1	11.4	4.4	20.0	0.7
Linefit	97.9	80.0	87.9	6.0	1.4	7.6	0.4
Patchwork	92.4	93.0	92.6	15.7	4.0	3.7	1.0
Patchwork++	93.3	93.7	93.5	13.2	3.8	4.0	1.6
R_GPF	67.9	95.3	78.6	17.4	12.4	2.0	1.1

TABLE III : Average performance (%) across all sequences and Runtime (Rt.) in milliseconds.

3D Point Cloud Segmentation

3D point cloud segmentation assigns semantic, instance, or part labels to points for scene understanding, mapping, robotics, and scientific measurement.

See all 70 papers

ACZ-GSeg: Adaptive Concentric Zone-based Two-stage Ground Segmentation for LiDAR Point Clouds
Jul 13, 2026Ge Zhang Chunyang Wang Bin Liu3D Point Cloud Segmentation Point Cloud Understanding
School of Optoelectronic Engineering, Xi'an University of Technology, Shaanxi, Xi'an, 710000
FugSeg: Fast Uncertainty-aware Ground Segmentation for 3D Point Cloud
May 9, 2026Yu Li, Volker Schwieger3D Point Cloud Segmentation Autonomous Driving Perception
Institute of Engineering Geodesy, University of Stuttgart, Stuttgart, Germany
TerraSeg: Self-Supervised Ground Segmentation for Any LiDAR
Mar 28, 2026Ted Lentsch, Santiago Montiel-Marín, Holger Caesar +13D Point Cloud Segmentation Self-Supervised Learning
Department of Cognitive Robotics, Delft University of Technology · Department of Electronics, University of Alcalá

Jul 13, 2026J/K move · Enter open · S save

ACZ-GSeg: Adaptive Concentric Zone-based Two-stage Ground Segmentation for LiDAR Point Clouds

Ge Zhang Chunyang Wang Bin Liu

School of Optoelectronic Engineering, Xi'an University of Technology, Shaanxi, Xi'an, 710000

3D Point Cloud Segmentation Point Cloud Understanding

Open paper AskPDF arXiv

Ground segmentation is a fundamental prerequisite for autonomous navigation, environmental perception, and object detection in ground mobile platforms. To address the under-segmentation of ground points caused by sparse long-range point clouds, ground undulations, and interference from non-ground structures in complex road scenarios, this paper proposes a two-stage ground segmentation method based on the Adaptive Concentric Zone Model. First, an Adaptive Concentric Zone Model is constructed to dynamically determine the number of sectors in each ring, thereby forming local zones with more balanced point distributions. Based on this model, a two-stage ground segmentation method is developed. In the coarse segmentation stage, a lowest-height seed constraint and height-decay weighting are introduced to establish a weighted principal component analysis plane fitting model, from which ground candidate points are extracted. In the fine segmentation stage, a reflectance intensity consistency constraint is employed to distinguish high-confidence ground points from uncertain points, and the uncertain points are further refined based on the local height stability of high-confidence neighborhoods. Experimental results show that the proposed method achieves Precision, Recall, and F1-score values of 99.12%, 96.24%, and 97.66% on the SemanticKITTI dataset, and 98.72%, 100.00%, and 99.36%, respectively, on a self-collected point cloud acquired using a RUBY-PLUS. The results demonstrate that the proposed method can effectively adapt to the range-dependent distribution characteristics of LiDAR point clouds, which are dense at near ranges and sparse at far ranges. It reduces the misclassification of non-ground points while maintaining ground point recall, thereby effectively improving the stability of ground segmentation.

Abstract

3D Point Cloud Segmentation

Explore similar work

ACZ-GSeg: Adaptive Concentric Zone-based Two-stage Ground Segmentation for LiDAR Point Clouds

FugSeg: Fast Uncertainty-aware Ground Segmentation for 3D Point Cloud

TerraSeg: Self-Supervised Ground Segmentation for Any LiDAR