Road Damage Detection Using Deep Convolutional Neural Networks

Road defects like potholes and cracking will decrease vehicle stability, expand braking distance and serious accidents that threaten the motorist. These damages are witnessed in some part of the road due to poor weather conditions, poor drainage and heavy traffic. Drivers previously used in car cameras or visual observation to keep an eye on the road surface, however, it is difficult to see both the road and the surface at the same time during the rush hour, at the places of low traffic, during bad weather etc. To fail over this challenge, the proposed system will examine the images of the road captured by the cameras automatically, and then group the surface as a smooth or rough area. It gives an instant alert to the driver in case there are any potholes or cracks. The automated real time detector cuts on driver manual monitoring, cuts on driver constant monitoring, and enhances more safety as the device ensures quick alert to the driver without necessarily having to be monitored by someone. Among the models that were selected to be used in the realtime driver-assistance, there is MobileNetV2, which is characterized by the fact that it is very fast and very accurate.