Experimental Analysis of Road Surface Deformation Quantification based on Unmanned Aerial Vehicle Images

Road surface deformation is a critical issue affecting transportation infrastructure and safety. This study presents an experimental analysis of road surface deformation quantification using Unmanned Aerial Vehicle (UAV) images. UAVs equipped with high-resolution cameras and LiDAR sensors were deployed to capture detailed images of road surfaces. A comprehensive preprocessing pipeline, including Structure from Motion (SfM), geometric correction, noise filtering, and contrast enhancement (CLAHE), was implemented to improve image quality. Advanced deep learning models, including Mask R-CNN, Vision Transformers (ViT), and Convolutional Neural Networks (CNNs), were employed for defect detection and classification. Mask R-CNN achieved the highest accuracy of 95.49%, outperforming CNN (88.67%) and LSTM-based models (80.44%). Feature extraction methods, such as Gray Level Co-occurrence Matrix (GLCM) and Wavelet Transform, were used to analyze texture properties and surface roughness. The integration of Geographic Information Systems (GIS) enabled heatmap generation and spatial mapping of deformation zones. The results demonstrate that UAV-based imaging combined with deep learning can provide an accurate, automated, and scalable solution for road surface monitoring. The proposed methodology can assist infrastructure authorities in early defect detection, predictive maintenance, and cost-effective road repairs.