Enhancing Infrastructure Monitoring: Pothole Detection in Road Images Using YOLOv8 and Open Datasets

Roadways are one of the main accesses for land transportation to be able to cross from one place to another. However, it's one of the public infrastructures that has the most damage problems to be maintained such as pothole problems. This paper aims to solve the pothole problem by making an AI model that can easily and quickly detect potholes so that the responsible staff and other road users can be informed to prevent greater damage and accidents that can occur due to potholes. The method for the development of a pothole detection system on roads is by using the YOLO (You Only Look Once) Object Detection based on deep learning algorithms and computer vision. This research was conducted by using a combination of 4 online open datasets of 2661 pothole images that were pre-processed by duplicate checking, removing low quality images, and adding pothole annotations. The dataset annotation is divided into 3 size categories, namely small, medium, and large potholes, which in each category has 500 images that have been labeled and augmented before. The training and validations result shows that the model is performing well in detecting road potholes with Precision, Recall, mAP50 over 70%, and mAP50-95 of 50%. Based on our experiments, we concluded that this object detection system is expected to increase the efficiency in detecting pothole damage scattered on road infrastructure spread over each region to be able to minimize further damage to road infrastructure and increase the safety of road users.