Скачать с ютуб Estimating winter road surface conditions - a deep learning based approach в хорошем качестве

Estimating winter road surface conditions - a deep learning based approach

Qian Xie: Winter road maintenance (WRM) plays a critical role in ensuring road safety and traffic mobility during the winter seasons. However, due to the vast areas that need to be monitored and serviced, WRM operations incur substantial costs pertaining to labor, material, equipment, etc. Thus, time and effort must be put forth to seek effective ways to minimize these costs while maintaining a high level of service. One approach is to obtain the road surface conditions (RSC) and their respective variations over space. Conventional approaches include a visual examination of camera images and on-site inspections by frequent road patrols. However, they are resource-intensive and thus impossible to be done in an efficient manner. To address this issue, this study aims at improving the efficiency of WRM operations by developing a deep learning (DL) model for automating the process of RSC image recognition. As a result, over 10,000 images were collected and labelled prior to model development. With training and validation accuracy reaching 99.89% and 94.62% respectively, the developed DL model is confirmed to be robust; therefore, it can contribute to the development of a real-time RSC monitoring solution that automatically generates descriptive RSC information in terms of snow coverage by using vehicle-mounted cameras.