Abstract: A detection system for road surface slippery condition is proposed by using the road surface wet state identification method based on transfer learning and the segmentation method for road water area based on full-resolution residual network to identify the road surface slippery condition and segment road water area. First, the proposed detection system is designed to identify four types of asphalt road surface conditions, including dry, wet, puddle, and waterlogged. Second, according to the waterlogged road surface identified by the detection system, the water area of the waterlogged road is segmented. Finally, the experiment is performed under various weather conditions. The results show that the designed detection system for road surface slippery condition can accurately, effectively, and real-time detect the asphalt road surface conditions under different illumination variations and various climate conditions.

Key words: Deep learning; Image processing; Road recognition; Puddle detection

摘要： 针对路面湿滑状态识别和积水区域分割问题，结合迁移学习的路面湿滑状态识别方法 和全分辨率残差网络的路面积水区域分割方法，设计一种路面湿滑状态检测系统。所设计的 检测系统对路面湿滑状态进行识别，包括干燥、潮湿、积水和积水淹没 4 种沥青路面，针对 检测系统识别出的积水路面，进一步对路面中积水区域进行分割，并在多种天气条件下进行 路面湿滑状态检测试验。试验结果表明，所设计路面湿滑状态检测系统能准确、有效且实时 地对不同光照和气候下的沥青路面湿滑状态进行检测。

关键词: 深度学习；图像处理；路面识别；积水检测