<p>This research addresses the issue of potholes as a significant threat to road safety and to vehicular integrity. It proposes a robust methodology for automatic pothole detection using the YOLOv5 deep learning framework, enriched with dataset optimization techniques made possible by Roboflow. The goal is to have very high detection accuracy for potholes, reaching a mean Average Precision of about 90%, for both static images and real-time video streams, by paying close attention to dataset preparation and fine-tuning the hyperparameters. The dataset undergoes extensive preprocessing, including annotation, data augmentation to simulate different environmental conditions, and normalization for increasing model generalizability. The different variants of YOLOv5, namely n, s, m, l, and x, are tested based on their precision, recall, F1-score, and map through standard evaluation metrics. The results indicate high performance of the system in detecting potholes on different road surfaces and lighting conditions. The value of this work is in pushing the edge of intelligent transportation systems by providing a scalable and deployable solution for the automation of road-condition monitoring, with the potential improvements in efficiency and safety in the maintenance of roads.</p>