Data Augmentation Strategies for Robust Lane Marking Detection

TL;DR

This paper tackles the domain shift problem in lane marking detection when deploying to non-standard camera viewpoints, such as side-mounted cameras. It introduces a generative AI–based data augmentation pipeline that combines geometric perspective transformation, AI-based inpainting, and vehicle body overlays to simulate deployment-like views and improve generalization. The approach is validated on two state-of-the-art lane detectors, SCNN and UFLDv2, showing substantial gains in precision, recall, and F1 over pre-trained baselines, particularly under challenging illumination and occlusion conditions. The proposed framework bridges the gap between public dashcam datasets and deployment scenarios, offering a scalable strategy to enhance lane-keeping reliability in pilot autonomous driving systems, while noting limitations to daytime data and outlining directions for future multimodal and self-supervised enhancements.

Abstract

Robust lane detection is essential for advanced driver assistance and autonomous driving, yet models trained on public datasets such as CULane often fail to generalise across different camera viewpoints. This paper addresses the challenge of domain shift for side-mounted cameras used in lane-wheel monitoring by introducing a generative AI-based data enhancement pipeline. The approach combines geometric perspective transformation, AI-driven inpainting, and vehicle body overlays to simulate deployment-specific viewpoints while preserving lane continuity. We evaluated the effectiveness of the proposed augmentation in two state-of-the-art models, SCNN and UFLDv2. With the augmented data trained, both models show improved robustness to different conditions, including shadows. The experimental results demonstrate gains in precision, recall, and F1 score compared to the pre-trained model. By bridging the gap between widely available datasets and deployment-specific scenarios, our method provides a scalable and practical framework to improve the reliability of lane detection in a pilot deployment scenario.

Figures (8)

• Figure 1: Overview of the proposed data augmentation pipeline. The process transforms CULane images into simulated local-domain data by applying geometric transformations, AI-based inpainting, and vehicle body overlays, generating training samples that approximate the target deployment perspective .
• Figure 2: Reference image captured from the locally mounted side-view camera. This setup provides a viewpoint oriented toward the wheels and adjacent lane markings, used as the target domain for domain adaptation experiments.
• Figure 3: Image comparison at different processing stages (top left: CULane original image, top right: university recording original image, bottom left: CULane after perspective transformation, bottom right: CULane after AI processing with lane line annotations)
• Figure 4: Binary mask used to isolate and remove the car hood region in CULane frames. This mask ensures consistent inpainting across sequences while preserving the surrounding lane context.
• Figure 5: Examples of stable–diffusion–based inpainting applied for vehicle hood removal. (a) illustrates a failure case with unnatural blending and texture artefacts. (b) shows a successful reconstruction of the occluded region, yielding plausible road surface and lane continuity.