Pothole Detection Using YOLOv3 Model

When it comes to moving people and goods around a country, road transports are like the heart and arteries. It is essential that roads need be kept in good shape at all times. Most of the times, potholes on the road surface causes major damages and they must be repaired for maintaining the roads in usable condition. Countries like India, where roads span millions of kilometers and require constant upkeep, it may be difficult to locate potholes. Since this is the case, rapid, accurate, and real-time automated pothole detection is essential. The primary motivation of this proposed work is to explain the pothole prediction models that are created using object detection algorithm called YOLOv3. The YOLOv3 model is trained on a dataset of potholes, and the results are analyzed and compared with variants of YOLO by computing the model's precision, recall, and mAP. When compared to other models, the mAP of this one is improved while the computing cost is reduced significantly.