Abstract

Due to the wide distribution of crosswalks over the road nets, the finding of impaired crosswalk marks is usually long-time delayed, which may put crosswalk pedestrians into danger. To reduce the repairing cost and improve the finding speed of damaged crosswalks, this paper uses remote sensing images to automatically detect crosswalks. The detection results can be used for further examination of crosswalks. However, the detection of crosswalks from remote sensing images suffers from serious interferes of many other kinds of ground targets. Besides, there are rare openly available datasets for the research of crosswalk detection from remote sensing images. To conquer the above problems, this study provides an openly available dataset for the research of crosswalk detection. To improve the robustness, we propose a crosswalk detection framework which uses a U-Net based road area guidance. First, we use CNN models to detect crosswalks. Then, we use U-Net to extract potential road areas. Third, we propose a mixture classification strategy which combines the detection confidence and potential road area guidance for final crosswalk detection. Experimental results show that the road area guidance for crosswalks’ detection is effective and can improve the detection performance.

摘要

由于人行横道在路网上的分布广泛，通常会长时间延迟发现受损的人行横道标记，这可能会使人行横道的行人面临危险。为了降低维修成本，提高受损人行横道的发现速度，本文采用遥感图像自动检测人行横道。检测结果可用于人行横道的进一步检查。然而，从遥感图像检测人行横道受到许多其他种类的地面目标的严重干扰。此外，很少有公开可用的数据集可用于遥感图像的人行横道检测研究。为了解决上述问题，本研究为人行横道检测的研究提供了一个公开可用的数据集。为了提高鲁棒性，我们提出了一个人行横道检测框架，该框架使用基于U-Net的道路区域指南。首先，我们使用CNN模型来检测人行横道。然后，我们使用U-Net提取潜在的道路区域。第三，我们提出了一种混合分类策略，将检测置信度和潜在的道路区域引导相结合，以进行最终的人行横道检测。实验结果表明，人行横道检测的道路引导是有效的，可以提高检测性能。