A Pothole Detection Using VGG16

Abstract

A pothole is a flaw that can be discovered on the road surface and it is one of the major contributions to the road accident. The impact of a vehicle on a potholed road is not just making the ride uncomfortable. It can damage the vehicle’s suspension system as well as the wheel of the vehicle, resulting in costly repair. Therefore, a regular road maintenance activity and assessment are very important to ensure that it is safe to be used. However, due to the limited number of expensive inspection vehicles, the inspection is performed manually. In this study, we present a mobile pothole detection system, namely HOLETRACKER using VGG16, a deep learning model architecture. The built model is trained using a collection of images taken from Kaggle and Internet in a variety of settings. The experiment used 739 numbers of training images and 144 numbers of testing images. The experimental result achieved the accuracy level rate at 90%. This paper also presents the development of two versions of the HOLETRACKER system, the mobile and web application that can be used by the public users and authorities. With the HOLETRACKER system, people can make a complaint of potholes via their mobile phone at anytime and anywhere. The validation checking of the potholed and location tracking through the GPS are the two main features provided by the system that will be performed before the information reaches the authorities for immediate action. The system is a cost-effective solution as an alternative to the manual pothole inspection management in facilitating the authorities as a measure to reduce accidents caused by potholes.