Learning to Automatically Catch Potholes in Worldwide Road Scene Images

Among several road hazards that are present in any paved way the world, potholes one of most annoying and also involving higher maintenance costs. There exists an increasing interest on automated detection these enabled by technological research progress. Our work tackled challenge pothole from images real world scenes. The main novelty resides application latest progress AI to learn visual appearance potholes. We built a large dataset with annotations. They contained scenes different cities taken cameras, vehicles viewpoints under varied environmental conditions. Then, we fine-tuned four object models based Faster R-CNN SSD deep neural networks. achieved high average precision detector was tested Nvidia DrivePX2 platform GPGPU capability, which can be embedded vehicles. Moreover, it deployed vehicle notify detected given IoT as part AUTOPILOT H2020 project.