Detection of Roads Potholes using YOLOv4

Road connectivity is most important for Developing Nations. A great challenge is to detect road damage which manually causes huge costs by putting different resources, such as instruments (technical devices). The need is to reduce the cost for governing bodies such that their inspection and damage road report should not be delayed. Road damage is of various types, which is an even regular process for the state to allocate budget in all means. The most efficient way which was seen by recent research is to utilize the latest object detection algorithm. A country like India where transportation is a lifeline for the economy. It has seen that most of the road damages arises due to potholes. Cost effective approach has to be considered to detect potholes. CNN has ability to extract the relevant features from image or video based datasets. In this paper, pothole image dataset annotated and trained using YOLOv4 and results were evaluated based on recall, preciosion and mAP. The model was validated with different set of images and videos based on potholes.