Abstract:

This research introduces a new method for automatically detecting road damage using deep learning algorithms and photos taken by Unmanned Aerial Vehicles (UAVs). In order to keep traffic safe and sustainable, road infrastructure must be regularly maintained. On the other hand, gathering information on road damage by hand may be dangerous and time-consuming. We thus suggest enhancing road damage detection with the use of UAVs and AI technology. For the purpose of object recognition and localization in UAV photos, our suggested method employs three algorithms: YOLOv4, YOLOv5, and YOLOv7. We put these algorithms through their paces using two datasets: one from Spain and one from China, the RDD2022. Our method successfully achieves 59.9% mean average accuracy mAP@.5 for the YOLOv5 version, 65.70% for a YOLOv5 model with a Transformer Prediction Head, and 73.20% for the YOLOv7 version, according to the testing data. These findings open the door to further study into the use of unmanned aerial vehicles (UAVs) and deep learning for automated road damage identification.

Description:

.