Answers are grounded in this paper and cite their sources.
Road infrastructure sustainability and public safety require monitoring and detection of potholes. Traditional methods of human-based traffic monitoring and pothole inspections are expensive, time-consuming and prone to human error. In this paper the usability of the YOLOv12 deep learning model to perform automated pothole detection in real-world conditions was examined. The authors generated a custom dataset consisting of 65 annotated video frames showcasing puddles or potholes which was used to train and validate the YOLOv12-based pothole detection model to determine its ability to accurately detect potholes and puddles. The YOLO series of models (you only look once) has shown success in real-time object detection in previous versions and YOLOv12 model is an advanced model with better detection capabilities. In this study, we have utilized the contextual workflow of preparing and preprocessing dataset, manual data annotation, and then training the YOLOv12 model to improve pothole detection. The experiments were analyzed using precision-recall curves, confusion matrices and F1-scores demonstrated YOLOv12's overall performance had high detection rates with minimal false positives. Visual confirmation of bounding box predictions provided additional assurance of accuracy and reliability in the model predictions. Overall the results demonstrate YOLOv12 is a promising solution for automated pothole detection with opportunities to reduce inspection expenses and add efficiency to road maintenance. Future work will evaluate scalability, possibilities in conjunction with sensor technologies and deploying for real-time applications on edge devices. The work contributes towards continued contributions to computer vision methods for road condition monitoring and supports growing transitions to smart infrastructure management systems.
No takes yet. Share an insight, caveat, or question.
Mohammed Basim Omar (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: