Identification of Potholes using Yolov4 and Darknet

Potholes and other pavement distresses are a matter of concern. Potholes not only cause harm to the vehicles but also endanger the lives of the people driving the vehicles. These potholes not only cause accidents that take lives but also cause minor back injuries that can become serious over a period of time. In India, as reported by the Ministry of Road Transport and Highways (MoRTH) around 5000 passengers are killed every year in road accidents caused by potholes. Pothole identification has been dependent upon manual detection, which is neither cost-effective nor time efficient. In this research study, we aim to accurately detect potholes with the help of two open-source datasets. We also intend to explore and compare various deep learning models to determine the optimum approach to detecting potholes in this work. After randomly mixing datasets to produce a combined dataset, we implemented variations of the YOLO model to improve efficiency.