SnowyLane: Robust Lane Detection on Snow-covered Rural Roads Using Infrastructural Elements

TL;DR

This work tackles lane detection in snow-covered rural roads by leveraging infrastructural delineators and Bezier-curve modeling, complemented by a large synthetic SnowyLane dataset containing $80{,}000$ frames with varying snow coverage. It introduces two robust approaches: (1) direct lane detection trained with the SnowyLane adverse-weather data, and (2) a delineator-based method that detects roadside posts and fits 3D Bezier curves to represent lane boundaries, including logic for lane counting and multi-lane handling. Experiments show that winter-data training substantially improves 2D and, depending on the detector, 3D accuracy and lane-safety metrics, while the delineator-based approach yields notably higher 3D accuracy and safety (LSM) than direct methods, with real-time performance on both camera- and LiDAR-based detectors. The dataset and methods together provide a strong foundation for all-weather autonomous driving and pave the way for extended evaluations in rain, fog, and beyond, with practical implications for robust, safe rural-road navigation.

Abstract

Lane detection for autonomous driving in snow-covered environments remains a major challenge due to the frequent absence or occlusion of lane markings. In this paper, we present a novel, robust and realtime capable approach that bypasses the reliance on traditional lane markings by detecting roadside features,specifically vertical roadside posts called delineators, as indirect lane indicators. Our method first perceives these posts, then fits a smooth lane trajectory using a parameterized Bezier curve model, leveraging spatial consistency and road geometry. To support training and evaluation in these challenging scenarios, we introduce SnowyLane, a new synthetic dataset containing 80,000 annotated frames capture winter driving conditions, with varying snow coverage, and lighting conditions. Compared to state-of-the-art lane detection systems, our approach demonstrates significantly improved robustness in adverse weather, particularly in cases with heavy snow occlusion. This work establishes a strong foundation for reliable lane detection in winter scenarios and contributes a valuable resource for future research in all-weather autonomous driving. The dataset is available at https://ekut-es.github.io/snowy-lane

Figures (6)

• Figure 1: Exemplary scene in a rural environment with partially snow-covered road which can be challenging for lane detection
• Figure 2: Different snow levels incorporated in the SnowyLane dataset. Low snow (left), medium snow (mid), and high snow (right)
• Figure 3: Example image from the summer weather condition SnowyLane subset with delineators
• Figure 4: Road layout of German rural roads. Image adapted from ral. Dimensions in mm
• Figure 5: Exemplary predictions of direct approach with (a) SCNN, (b) RESA and (c) UFLD as well as the delineator-based approach with (d) Faster R-CNN, (e) Voxel R-CNN and (f) PV-RCNN.