Road Surface Classification and Subsequent Pothole Detection Using Deep Learning
Ravi Agrawal, Yash Chhadva, Sanjana Addagarla, Sheetal Chaudhari
Abstract
Potholes are a huge road safety concern. Using a Convolutional Neural Network (CNN) model to detect and identify potholes will save manual labour, time and taxpayer money. The proposed model will take in images of the road and efficiently will classify which category the road belongs to, i.e, Asphalt, Paved or Unpaved and then the respective detection models will further detect the potholes present in that image. An application built on this model will also be able to alert pedestrians of existing nearby potholes to prevent any accidents or discomfort. This research shows that the model can detect potholes with an accuracy of 88% and road types accurately with an accuracy of 96%.