USING UAV PHOTOS AND DEEP LEARNING TO AUTOMATICALLY DETECT ROAD DAMAGE

Abstract

Using deep learning techniques and images from unmanned aerial vehicles (UAVs), this study presents a novel automated technique for identifying road damage. Road infrastructure upkeep is necessary for a transportation system to be both sustainable and safe. However, collecting data on road damage manually could be time-consuming and hazardous for humans. Therefore, in order to significantly improve the precision and effectiveness of road damage diagnosis, we propose utilizing UAVs in conjunction with Artificial Intelligence (AI) technology. Our proposed method for object recognition and localization in UAV pictures uses three algorithms: YOLOv4, YOLOv5, and YOLOv7. To train and assess these systems, we combined the RDD2022 dataset from China with a road dataset from Spain. The experimental results confirm the efficacy of our approach, achieving mean average precision ([email protected]) of 59.9% for the YOLOv5 version, 65.70% for a YOLOv5 model with a Transformer Prediction Head, and 73.20% for the YOLOv7 version. These results demonstrate the potential of automated road damage identification using UAVs and deep learning and pave the way for further research in this field.