Pothole Detection from an Enhanced Aerial Image Using CNN Model

The presence of potholes in road surface can cause fatal road accidents. This can even lead to mishap of vehicles and causing property damages. An automated system of detecting potholes in road surfaces becomes one of the utility services in any driver assistance system. The proposed methodology aims at providing a warning sign to the driver through driver assistance system regarding the presence of pothole along with its risk category. Based on the road surface image captured by the front camera fixed on the vehicle, the proposed convolutional neural network (CNN) model identifies and classifies the pothole. The captured image is initially enhanced to reduce the complexity load for the CNN model and to improve the accuracy of the model. The proposed model is tested and validated over 1000 sample aerial images and it classifies the images with 91.13% accuracy. The driver assistance system with this service can alert the driver with the presence of pothole.