GHySSeg: Real-Time Geometric Ground and Road Segmentation for Hybrid Solid-State LiDAR
Abstract
Hybrid solid-state LiDAR sensors are making their way into production vehicles, yet the data they produce differs structurally from that of the established spinning mechanical sensors behind today's autonomous-driving datasets: it is ring-free, dense, cluttered, and concentrated in a frontal field of view. The classical geometric methods that autonomous vehicles rely on for drivable space computation and navigation assume the ordered ring-and-column structure of a 360-degree spinning LiDAR and degrade or fail on this data, while learned alternatives are computationally heavy, hard to certify, and trained almost exclusively on spinning-LiDAR data. We demonstrate that the same dense, per-curve geometry is instead an opportunity to close the divide between the industry's preferred sensors and autonomous-driving research. We present GHySSeg, a purely geometric method that segments ground and drivable road from a single hybrid solid-state LiDAR frame by tracing each scan curve and reading two geometric signatures, a step in height and a fold-back in range, with no training and no graphics processing unit. To evaluate it, we release HySoLiD-Drive, the first public benchmark dedicated to ground and road segmentation on an automotive hybrid solid-state LiDAR, comprising 100 manually labeled and 7000 unlabeled frames recorded by two vehicles in real driving. On this benchmark, GHySSeg surpasses both geometric and learned state-of-the-art baselines across ground and road segmentation, reaching an intersection-over-union of 0.907 for road and 0.896 for ground at over 0.96 precision, while running in real time at about 22 milliseconds per frame on a single core.
Figures & tables
| Dataset | Sensor(s) | Type | Environment | Benchmark | Ground | Road |
|---|---|---|---|---|---|---|
| SemanticKITTI [ 8 ] | Velodyne HDL-64E | Mechanical | Urban driving | ✓ | ✓ | ✓ |
| nuScenes [ 33 ] | Velodyne HDL-32E | Mechanical | Urban driving | ✓ | ✓ | ✓ |
| Waymo Open [ 34 ] | 1 top + 4 perimeter | Mechanical | Urban driving | ✓ | ✓ | ✓ |
| PandaSet [ 35 ] | Pandar64 + PandarGT | Mechanical + Hybrid SS | Urban driving | ✗ | ✓ | ✓ |
| SimoSet [ 36 ] | Forward hybrid SS | Hybrid SS | Urban driving | ✓ | ✗ | ✗ |
| S.MID [ 37 ] | Livox Mid-360 | Hybrid SS | Industrial | ✓ | ✓ | ✓ |
| ATLAS ( ) | Kalipso ( ) | All ( ) | |||||||||||
| Method | Type | IoU | P | R | F1 | IoU | P | R | F1 | IoU | P | R | F1 |
| Ground segmentation | |||||||||||||
| GHySSeg (ours) | Geometric | 0.882 | 0.963 | 0.914 | 0.936 | 0.911 | 0.960 | 0.947 | 0.953 | 0.896 | 0.961 | 0.931 | 0.944 |
| Patchwork++ [ 17 ] | Geometric | 0.874 | 0.917 | 0.948 | 0.931 | 0.879 | 0.931 | 0.940 | 0.934 | 0.877 | 0.924 | 0.944 | 0.933 |
| CSF [ 18 ] | Geometric | 0.837 | 0.855 | 0.979 | 0.903 | 0.863 | 0.863 | 0.999 | 0.925 | 0.850 | 0.859 | 0.989 | 0.914 |
| GndNet [ 28 ] | Learned | 0.747 | 0.751 | 0.993 | 0.851 | 0.724 | 0.737 | 0.978 | 0.835 | 0.736 | 0.744 | 0.985 | 0.843 |
| IoU per scenario bucket | All ( ) | ||||||||||
| Configuration | Crossw. | Roundab. | Sharp t. | Ramp | Intersec. | Parking | Curve | Straight | IoU | P | R |
| Ground segmentation | |||||||||||
| Seeded region grow | 0.869 | 0.852 | 0.799 | 0.732 | 0.851 | 0.851 | 0.818 | 0.808 | 0.825 | 0.860 | 0.953 |
| + vertical-stack gate | 0.912 | 0.894 | 0.875 | 0.781 | 0.898 | 0.924 | 0.883 | 0.879 | 0.885 | 0.934 | 0.942 |
| + rank re-seed | 0.909 | 0.891 | 0.874 | 0.776 | 0.900 | 0.922 | 0.876 | 0.878 | 0.883 | 0.922 | 0.954 |
| + protrusion guard (full) | 0.910 | 0.900 | 0.892 | 0.825 | 0.916 | 0.923 | 0.878 | 0.895 | 0.896 | 0.961 | 0.931 |
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 79 papers