Optimizing drivable area detection for autonomous vehicles through shadow and road edge line analysis in U-net-based models

Engineering notes

Key topics: autonomous driving, autonomous vehicle. See the paper for implementation details and experimental results.

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

Reliable drivable area detection is essential for autonomous vehicle operation. However, built environments are often confused with other objects that share similar local features because of these environments’ variability and complexity. Utilizing an improved U-Net model, this research systematically evaluates how road shadows and edge markings affect detection accuracy while exploring built environment optimization. Experimental results demonstrate that strong shadows reduce drivable area detection accuracy from 95.9% to 88.2% under consistent road conditions. Complex local contrast changes caused by shadows lead to redundant elements corresponding to shadow boundaries during the segmentation process. Detection accuracy dropped from 95.6% to 91.4% due to missing or worn road edge lines. Furthermore, in areas with missing edge lines, the accuracy of detecting drivable areas is closely related to the color contrast with the surroundings. This study suggests that future road designs for autonomous driving should minimize environmental elements that create strong shadows, such as tall trees or buildings. In addition, autonomous driving roads should have clear edges and high color contrast between drivable areas and the surrounding nondrivable areas.

Links and sources

Comments