Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Robust Lane Detection with Wavelet-Enhanced Context Modeling and Adaptive Sampling

Kunyang Li, Ming Hou

TL;DR

This work tackles robust lane detection for autonomous driving under adverse conditions such as occlusion, extreme illumination, and complex curves. It introduces a Wavelet-Enhanced Feature Pyramid Network (WE-FPN) to improve global context while preserving local lane details, an adaptive preprocessing module to boost lane visibility in poor lighting, and an attention-guided sampling strategy to emphasize long-range geometric cues. Comprehensive experiments on CULane and TuSimple demonstrate state-of-the-art performance, with clear ablations showing gains from data preprocessing, attention sampling, and WE-FPN, particularly for curved and low-light scenarios. The method achieves robust, real-time lane detection that can enhance safety-critical perception in real-world driving, with potential for extension to multi-modal sensor fusion and dynamic lane topology prediction.

Abstract

Lane detection is critical for autonomous driving and ad-vanced driver assistance systems (ADAS). While recent methods like CLRNet achieve strong performance, they struggle under adverse con-ditions such as extreme weather, illumination changes, occlusions, and complex curves. We propose a Wavelet-Enhanced Feature Pyramid Net-work (WE-FPN) to address these challenges. A wavelet-based non-local block is integrated before the feature pyramid to improve global context modeling, especially for occluded and curved lanes. Additionally, we de-sign an adaptive preprocessing module to enhance lane visibility under poor lighting. An attention-guided sampling strategy further reffnes spa-tial features, boosting accuracy on distant and curved lanes. Experiments on CULane and TuSimple demonstrate that our approach signiffcantly outperforms baselines in challenging scenarios, achieving better robust-ness and accuracy in real-world driving conditions.

Robust Lane Detection with Wavelet-Enhanced Context Modeling and Adaptive Sampling

TL;DR

This work tackles robust lane detection for autonomous driving under adverse conditions such as occlusion, extreme illumination, and complex curves. It introduces a Wavelet-Enhanced Feature Pyramid Network (WE-FPN) to improve global context while preserving local lane details, an adaptive preprocessing module to boost lane visibility in poor lighting, and an attention-guided sampling strategy to emphasize long-range geometric cues. Comprehensive experiments on CULane and TuSimple demonstrate state-of-the-art performance, with clear ablations showing gains from data preprocessing, attention sampling, and WE-FPN, particularly for curved and low-light scenarios. The method achieves robust, real-time lane detection that can enhance safety-critical perception in real-world driving, with potential for extension to multi-modal sensor fusion and dynamic lane topology prediction.

Abstract

Lane detection is critical for autonomous driving and ad-vanced driver assistance systems (ADAS). While recent methods like CLRNet achieve strong performance, they struggle under adverse con-ditions such as extreme weather, illumination changes, occlusions, and complex curves. We propose a Wavelet-Enhanced Feature Pyramid Net-work (WE-FPN) to address these challenges. A wavelet-based non-local block is integrated before the feature pyramid to improve global context modeling, especially for occluded and curved lanes. Additionally, we de-sign an adaptive preprocessing module to enhance lane visibility under poor lighting. An attention-guided sampling strategy further reffnes spa-tial features, boosting accuracy on distant and curved lanes. Experiments on CULane and TuSimple demonstrate that our approach signiffcantly outperforms baselines in challenging scenarios, achieving better robust-ness and accuracy in real-world driving conditions.

Paper Structure

This paper contains 27 sections, 3 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Deep Learning-based Lane Detection
  4. Feature Enhancement and Context Modeling
  5. Approach
  6. Data Preprocessing
  7. Motivation.
  8. Wavelet-Enhanced FPN (WE-FPN)
  9. Attention Sampling
  10. Motivation.
  11. The formula for this Attention Sampling is:
  12. Training and Infercence Details
  13. Positive samples selection.
  14. Training Loss.
  15. Inference.
  16. ...and 12 more sections

Figures (4)

  • Figure 1: Illustrations of hard cases for lane detection. (a) The case that lane is blurred by the extreme lighting condition. (b) The rode with complex paths.
  • Figure 2: Overview of the proposed our. (a) The network generates feature maps from WE-FPN structure. Subsequently, each lane prior will be refined from high-level features to low-level features. (b) The input layers feed into internal layers integrated with positional non-local blocks to capture spatial context. (c) The internal layers connect to output layers that pass through Attention Sampling.
  • Figure 3: Visual depiction comparing Attention Sampling (red dots) versus uniform sampling (blue dots).
  • Figure 4: Ablation study on the weighted combination of WE-FPN and FPN, where the contribution of WE-FPN is controlled by the weighting factor $\alpha$ (ranging from 0.1 to 0.9). The results are computed as WE-FPN*$\alpha$ + FPN*(1-$\alpha$), demonstrating the impact of varying the influence of WE-FPN on model performance across different scenarios.