A Pothole Detection and Mapping Approach Using Yolo-Based Federated Learning for Transportation Solutions in Smart Cities

Abstract

In today's rapidly growing cities, traffic management and the condition of road infrastructure pose a significant challenge in terms of both efficiency and safety. The most important of these challenges is the potholes that form on city roads. Potholes negatively affect traffic, leading to reduced transportation efficiency and accidents. Autonomous vehicles in particular are greatly affected by potholes. Detection of potholes usually depends on citizens alerting the authorities. However, this method is not always effective as most people may be insensitive. An innovative solution to these problems has been developed using the smart city concept and deep learning methods. In this study, a federated deep learning-based approach was developed for pothole detection and mapping using images obtained from vehicle cameras in order to increase the efficiency of smart cities. In this way, the detected potholes were quickly fixed by the authorities, drivers were warned in advance and transportation safety was increased. Damage to vehicles due to poor road conditions was reduced, and in cases where potholes would negatively affect traffic, alternative routes were suggested, and time loss was minimized. In our study, a federated object detection model developed with the YOLO model was designed and real-time detection and mapping of potholes was provided. Thus, an important contribution has been made to the literature. The images to be used were obtained through cameras in the vehicles. By using the federated deep learning approach, bottlenecks that may occur in network traffic were prevented by prioritizing the confidentiality and security of the data coming from the vehicles. In this article, some federated deep learning models used for pothole detection and basic studies in the literature are examined. © The Institution of Engineering & Technology 2024.