HSDF-Lane: Height-Aligned Signed Distance Field with Semantic Lane Prior for 3D Lane Detection
Abstract
Monocular 3D lane detection plays a critical role in autonomous driving, yet recovering reliable 3D geometry from a single image remains challenging due to inherent depth ambiguity. Prior methods project image features into Bird's-Eye-View (BEV) space under a flat-ground assumption, causing geometric distortion on real-world roads. Recent methods instead predict explicit height maps to capture non-planar surfaces, but still rely on sparse anchor-based regression and exploit the recovered geometry merely for spatial transformation rather than semantic understanding. To overcome these limitations, we propose HSDF-Lane, which implicitly models the road surface as a Height-aligned Signed Distance Field (HSDF) over a densely sampled 3D feature volume. Through differentiable rendering, the HSDF jointly produces an accurate height map and surface-aligned features. We further introduce Lane-aware Semantic Positional Encoding (LSPE), which injects a lane-existence prior derived from the surface-aligned features into the transformer queries, coupling geometric structure with semantic guidance. Extensive experiments on the OpenLane benchmark show that HSDF-Lane achieves state-of-the-art performance in both 3D lane detection and height map estimation.
Results
Heightmap Estimation Demo Comparison of heightmap estimation between HSDF-Lane and SC-Lane (baseline) in steep uphill and downhill scenarios. The top section visualizes the estimated 3D heightmaps in world coordinates, while the bottom section displays the corresponding HSDF values and rendered heights in ego coordinates. The results demonstrate the robustness and effectiveness of HSDF modeling in handling abrupt elevation changes.
3D Lane Detection Demo Comparison of 3D lane detection performance between HSDF-Lane and SC-Lane (baseline) during sharp curve transitions. HSDF-Lane achieves superior detection accuracy in highly curved scenes. Furthermore, attention map visualizations reveal that while the baseline exhibits scattered attention in irrelevant regions, HSDF-Lane effectively localizes attention strictly around the lane structures, facilitated by LSPE.
Quantitative Results
| Method | F1-Score(%)↑ | X-error near(m)↓ |
X-error far(m)↓ |
Z-error near(m)↓ |
Z-error far(m)↓ |
|---|---|---|---|---|---|
| PersFormer | 50.5 | 0.485 | 0.553 | 0.364 | 0.431 |
| BEV-LaneDet | 58.4 | 0.309 | 0.659 | 0.244 | 0.631 |
| LATR | 61.9 | 0.219 | 0.259 | 0.075 | 0.104 |
| PVALane | 63.4 | 0.226 | 0.257 | 0.093 | 0.119 |
| GroupLane | 64.1 | 0.320 | 0.441 | 0.233 | 0.402 |
| HeightLane | 62.7 | 0.240 | 0.266 | 0.116 | 0.165 |
| SC-Lane | 64.3 | 0.227 | 0.251 | 0.088 | 0.128 |
| Rethinking | 64.7 | 0.205 | 0.255 | 0.074 | 0.105 |
| GLane3D-B | 63.9 | 0.193 | 0.234 | 0.065 | 0.090 |
| SparseLaneSTP | 66.1 | 0.203 | 0.240 | 0.066 | 0.092 |
| HSDF-Lane (Ours) | 66.3 | 0.201 | 0.223 | 0.088 | 0.114 |
| HSDF-Lane† (Ours) | 66.9 | 0.186 | 0.226 | 0.084 | 0.114 |
BibTeX
@misc{boo2026hsdflaneheightalignedsigneddistance,
title={HSDF-Lane: Height-Aligned Signed Distance Field with Semantic Lane Prior for 3D Lane Detection},
author={Jiyong Boo and Byeongin Joung and Hyemin Yang and Kuk-Jin Yoon},
year={2026},
eprint={2606.31172},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2606.31172},
}