Author(s) :

Pillala Dhanarjuna.

Volume/Issue :

Abstract :

Auto-detection of road damage is achieved through deep learning using images absorbed by unmanned aircraft. Maintaining street infrastructure is of great importance for safe and sustainable transportation, but manual data collection is often labor-intensive and dangerous. To address this, we propose leveraging UAVs and AI to significantly improve detection efficiency and accuracy. Object detection and localization in UAV imagery utilizes yolov4, yolov5, and yolov7 algorithms, which have been trained on the rdd2022 dataset from China and a dataset of Spanish roads. YOLOv5 yielded 59.9% mAP@.5 whereas YOLOv5 equipped with a Transformer Prediction Head attained surprisingly high 65.70% mAP@.5 and YOLOv7 outperformed both achieving 73.20% mAP@.5. UAV-based deep learning systems show considerable promise for automated detection of road damage offering a robust foundation for future smart infrastructure research endeavors.

No. of Downloads :

0