Wu Junjie, Liu Wen, Maruyama Yoshihisa
Nippon Koei Co., Ltd., 5-4 Kojimachi, Chiyoda-ku, Tokyo 102-8539, Japan.
Graduate School of Engineering, Chiba University, Inage-ku, Chiba 263-8522, Japan.
Sensors (Basel). 2024 Dec 3;24(23):7724. doi: 10.3390/s24237724.
Road markings are vital to the infrastructure of roads, conveying extensive guidance and information to drivers and autonomous vehicles. However, road markings will inevitably wear out over time and impact traffic safety. At the same time, the inspection and maintenance of road markings is an enormous burden on human and economic resources. Considering this, we propose a road marking inspection system using computer vision and deep learning techniques with the aid of street view images captured by a regular digital camera mounted on a vehicle. The damage ratio of road markings was measured according to both the undamaged region and region of road markings using semantic segmentation, inverse perspective mapping, and image thresholding approaches. Furthermore, a road marking damage detector that uses the YOLOv11x model was developed based on the damage ratio of road markings. Finally, the mean average precision achieves 73.5%, showing that the proposed system successfully automates the inspection process for road markings. In addition, we introduce the Road Marking Damage Detection Dataset (RMDDD), which has been made publicly available to facilitate further research in this area.
道路标线对于道路基础设施至关重要，它向驾驶员和自动驾驶车辆传达广泛的引导和信息。然而，道路标线不可避免地会随着时间的推移而磨损，并影响交通安全。与此同时，道路标线的检查和维护对人力和经济资源来说是一个巨大的负担。考虑到这一点，我们借助安装在车辆上的普通数码相机拍摄的街景图像，提出了一种使用计算机视觉和深度学习技术的道路标线检测系统。使用语义分割、逆透视映射和图像阈值处理方法，根据道路标线的未损坏区域和损坏区域来测量道路标线的损坏率。此外，基于道路标线的损坏率，开发了一种使用YOLOv11x模型的道路标线损坏检测器。最后，平均精度达到了73.5%，表明所提出的系统成功地实现了道路标线检查过程的自动化。此外，我们还介绍了道路标线损坏检测数据集（RMDDD），该数据集已公开提供，以促进该领域的进一步研究。
