Fast Grid-Based Ground Segmentation of LiDAR Point Clouds Using Dual-Seed Expansion

Engineering notes

Key topics: autonomous driving, lidar, point cloud, kitti, perception. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Ground segmentation is an essential preprocessing step for LiDAR-based perception in autonomous driving, as it directly affects subsequent object clustering and obstacle detection. This paper proposes a fast grid-based ground segmentation method for 3D Light Detection and Ranging (LiDAR) point clouds using dual-seed expansion. The input point cloud is projected onto a two-dimensional Cartesian grid, where cell-wise height statistics, including minimum height, maximum height, mean height, and height variance, are computed. Initial ground candidates are selected by combining a global height-distribution-based seed and a LiDAR-mounting-height-based seed to reduce missed ground candidates in sloped or sparse regions. Ground cells are then expanded to nearby cells based on the local height difference. To reduce misclassification in cells containing both ground and non-ground points, a local height-band criterion is applied for point-level refinement. The proposed method was implemented in Robot Operating System 1 (ROS 1) and evaluated on the SemanticKITTI dataset. Experimental results showed an average accuracy of 96.34%, an F1-score of 97.11%, and a ground Intersection over Union (IoU) of 94.40%. The average processing time over SemanticKITTI sequences 00–10 was 12.15 ms per frame. These results demonstrate that the proposed method is suitable as a fast preprocessing module for LiDAR-based autonomous driving perception.

Links and sources

Comments