Real Time Pothole Detection using Machine Learning and Edge Computing

Authors:

Abstract
Vehicle owners in every region of the globe face a significant challenge in the form of potholes. They result in property damage to cars, lengthen the time it takes to get from place to place and present a threat to the safety of motorists, cyclists, and pedestrians. This paper suggests a refreshing approach for detecting potholes using machine learning and edge computing technologies. The YOLOv7 model architecture is employed for pothole detection by applying transfer learning techniques. On the edge Raspberry Pi device, the trained model is deployed. Our proposed system consists of a camera module that is connected to a Raspberry Pi. The camera is affixed to the vehicle's dashboard and transmits live footage of the road surface to the edge device. The Raspberry Pi conducts inference on the stream of road surface data captured by the camera, detecting and preserving images of identified potholes. The efficacy of the suggested method has been demonstrated through experimental findings, exhibiting a notable degree of precision and expediency in the identification of potholes. The amalgamation of machine learning and edge computing presents a pragmatic resolution that can be implemented on low-energy consumption devices, such as Raspberry Pi, thereby rendering it both scalable and economical for extensive deployment. The research introduces a pothole detection system that operates in near real-time. The system's implementation is expected to enhance proactive road maintenance, resulting in a reduction of infrastructure damage and an improvement in road safety.