Accurate Identification of Potholes on the Road using Federated Learning

In developing countries like Bangladesh maintaining roads properly is a challenging task. Potholes on the road can lead to various accidents putting the lives and property of drivers, passengers, and pedestrians at risk. Due to increasing the number of potholes on the roads, the number of accidents is climbing day by day. Therefore, continuous collection and up-gradation of road condition data with the most recent information are highly necessary. Identification of potholes can abet drivers to select the right path to avoid accidents or vehicle damage as well as assist the concerned Government department to take immediate measures to fill up the potholes for the benefit of the commuters. Concerning this problem, existing solutions in the literature are highly error-prone and time-consuming. On the other hand, no one gives concentration to renew the information on potholes in a little period. In this paper, we develop an immensely proper pothole identification system by using federated learning. The method recognizes potholes and produces a notification for drivers. Whenever a vehicle faces a pothole a response notification is transmitted to the central server, which then aggregates whole data and gets updated road conditions in every phase. The proposed system outclasses the latest works with which it is compared in a matter of accuracy.