Neumann, Felix (Siemens AG), Deroo, Frederik (Siemens AG), v. Wichert, Georg (Siemens AG), Burschka, Darius (Technical University Munich)

LiDAR Ground Segmentation with Gaussian Mixture Models

Scheduled for presentation during the Regular Session "S43b-Multi-Sensor Fusion and Perception for Robust Autonomous Driving" (FR-EA-T43), Friday, November 21, 2025, 14:30−14:50, Stradbroke

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Accurate ground segmentation is an important perception task, not only for estimating drivable surfaces, but also as a precursor for tasks such as object clustering or dynamic object segmentation and tracking in autonomous vehicles. Model-based methods have made substantial progress on this problem in recent times, but place their focus on generating a model of the ground, while modeling non-ground objects is not emphasized. However, modeling such objects can provide important additional information as evidence for non-ground points. We propose a Gaussian Mixture Model-based environment model to estimate the likelihood that local regions belong to the ground or non-ground objects, which can be queried at arbitrary positions in space. Additionally, we extend this approach to fuse information from multiple past sensor frames for more accurate ground estimation. We experimentally validate our approach on the SemanticKITTI dataset, where notably our single-frame configuration outperforms state-of-the-art multi-frame methods.