Deep Learning Model In Road Surface Condition Monitoring

Road is a crucial asset for public services. Therefore, it must be repaired as soon as possible to bring safety for the driver on the road. In recent years research in monitoring and identifying road surface conditions using images captured by the low-cost camera or smartphone cameras and deep learning can identify road surface conditions efficiently and quickly. The study aims to explore the performance of the deep learning model. We proposed a U-Net architecture for road extraction from Road Traversing Knowledge (RTK) dataset. The result has been compared with different automatic segmentation models, and the development of segmentation using U-Net showed a more precise Accuracy of 97.08%, Mean-IoU 0.364, and faster computing time than the other models.