Advances in Lane Detection: From Classical Methods to Transformer-Based Architectures

Engineering notes

This paper reviews state-of-the-art lane detection frameworks, tracing their evolution from classical geometric models (e.g., Hough Transform) to convolutional neural networks (CNNs), anchor-based models, and transformer-based architectures, with a focus on unstructured roads.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

Lane detection is a critical component of Advanced Driver Assistance Systems (ADAS) and autonomous navigation, especially in unstructured environments where lane markings are often degraded or missing. Ever-increasing driving complexity has shifted research from traditional image processing to deep learning approaches. This paper reviews state-of-the-art lane detection frameworks, tracing their evolution from classical geometric models (e.g., Hough Transform) to convolutional neural networks (CNNs), anchor-based models, and transformer-based architectures, with a focus on unstructured roads. The review covers traditional vision-based approaches, deep learning methods such as semantic segmentation, anchor-based models, and transformer architectures, while addressing challenges like illumination changes, inconsistent markings, and environmental noise. Applications range from commercial ADAS to research prototypes and simulation platforms.

Links and sources

• Official / arXiv page

Comments