Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

LOID: Lane Occlusion Inpainting and Detection for Enhanced Autonomous Driving Systems

Aayush Agrawal, Ashmitha Jaysi Sivakumar, Ibrahim Kaif, Chayan Banerjee

TL;DR

The paper tackles lane-detection under occlusion, proposing two approaches: aug-Segment, a data-augmentation–driven segmentation method, and LOID, a three-node Lane Occlusion Inpainting and Detection pipeline. aug-Segment yields ~12% IoU gains on the occlusion-augmented CULanes dataset but shows limited transfer to BDD100K. LOID delivers state-of-the-art performance on both augmented datasets, achieving IoU up to 0.533 on BDD100K and 0.302 on CULanes by detecting occlusions, inpainting occluded regions, and producing lane masks, with Dice scores significantly higher than baselines. The method achieves real-time capability (~25 FPS) and demonstrates robustness across datasets, though it has higher inference time than some baselines, motivating future work to merge detector and inpainting stages. Overall, LOID offers a practical, occlusion-resilient solution for lane detection in autonomous driving systems, with aug-Segment providing a lightweight alternative when occlusion scenarios are dataset-aligned.

Abstract

Accurate lane detection is essential for effective path planning and lane following in autonomous driving, especially in scenarios with significant occlusion from vehicles and pedestrians. Existing models often struggle under such conditions, leading to unreliable navigation and safety risks. We propose two innovative approaches to enhance lane detection in these challenging environments, each showing notable improvements over current methods. The first approach aug-Segment improves conventional lane detection models by augmenting the training dataset of CULanes with simulated occlusions and training a segmentation model. This method achieves a 12% improvement over a number of SOTA models on the CULanes dataset, demonstrating that enriched training data can better handle occlusions, however, since this model lacked robustness to certain settings, our main contribution is the second approach, LOID Lane Occlusion Inpainting and Detection. LOID introduces an advanced lane detection network that uses an image processing pipeline to identify and mask occlusions. It then employs inpainting models to reconstruct the road environment in the occluded areas. The enhanced image is processed by a lane detection algorithm, resulting in a 20% & 24% improvement over several SOTA models on the BDDK100 and CULanes datasets respectively, highlighting the effectiveness of this novel technique.

LOID: Lane Occlusion Inpainting and Detection for Enhanced Autonomous Driving Systems

TL;DR

The paper tackles lane-detection under occlusion, proposing two approaches: aug-Segment, a data-augmentation–driven segmentation method, and LOID, a three-node Lane Occlusion Inpainting and Detection pipeline. aug-Segment yields ~12% IoU gains on the occlusion-augmented CULanes dataset but shows limited transfer to BDD100K. LOID delivers state-of-the-art performance on both augmented datasets, achieving IoU up to 0.533 on BDD100K and 0.302 on CULanes by detecting occlusions, inpainting occluded regions, and producing lane masks, with Dice scores significantly higher than baselines. The method achieves real-time capability (~25 FPS) and demonstrates robustness across datasets, though it has higher inference time than some baselines, motivating future work to merge detector and inpainting stages. Overall, LOID offers a practical, occlusion-resilient solution for lane detection in autonomous driving systems, with aug-Segment providing a lightweight alternative when occlusion scenarios are dataset-aligned.

Abstract

Accurate lane detection is essential for effective path planning and lane following in autonomous driving, especially in scenarios with significant occlusion from vehicles and pedestrians. Existing models often struggle under such conditions, leading to unreliable navigation and safety risks. We propose two innovative approaches to enhance lane detection in these challenging environments, each showing notable improvements over current methods. The first approach aug-Segment improves conventional lane detection models by augmenting the training dataset of CULanes with simulated occlusions and training a segmentation model. This method achieves a 12% improvement over a number of SOTA models on the CULanes dataset, demonstrating that enriched training data can better handle occlusions, however, since this model lacked robustness to certain settings, our main contribution is the second approach, LOID Lane Occlusion Inpainting and Detection. LOID introduces an advanced lane detection network that uses an image processing pipeline to identify and mask occlusions. It then employs inpainting models to reconstruct the road environment in the occluded areas. The enhanced image is processed by a lane detection algorithm, resulting in a 20% & 24% improvement over several SOTA models on the BDDK100 and CULanes datasets respectively, highlighting the effectiveness of this novel technique.
Paper Structure (20 sections, 6 figures, 4 tables)

This paper contains 20 sections, 6 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Methodology
  4. Architecture Design -- aug-Segment
  5. Architecture Design -- LOID
  6. Detector Node
  7. Inpainting Node
  8. Segmentation
  9. Datasets & Evaluation
  10. Augmented CULanes Training Set
  11. Augmented Validation Set
  12. Evaluation
  13. Experiments and Results
  14. Experiments
  15. Results
  16. ...and 5 more sections

Figures (6)

  • Figure 1: Semantic Mask Generated by Different Models.
  • Figure 2: Overview of LOID Architecture. The detector node is responsible for identifying and classifying the occlusions. The inpainting node uses a Coarse and Refinement network to regenerate occlusion-free markings. Finally, a segmentation node is used to generate the final lane masks.
  • Figure 3: Lane Segmented Outputs from various models | BDD
  • Figure 4: Lane Segmented Outputs from various models | CULanes
  • Figure 5: Pipeline Outputs - BDD
  • ...and 1 more figures