AI-based Pothole Detection for Enhanced Road Safety: A Comprehensive Review of Architectures, Modal Fusion and Edge Deployment Strategies

The integrity of road infrastructure is a critical determinant of transportation. Road infrastructure quality is essential for safety and economic efficiency, but potholes are a global problem of traffic and environmental stress. Manual inspections are not reliable, expensive and dangerous. This paper reviews the moving towards AI-based pothole detection in the Intelligent Transportation Systems (ITS) field, which covers the sensing methods from smartphone-based vibration to the advanced 2D/3D computer vision. It focuses on the deep learning innovations (YOLOv5 to YOLOv11), as well as Vision Transformers and GANs for better detection and image enhancement. The study also investigates multi-modal sensor fusion (RGB, thermal, LiDAR) for addressing issues such as low light and wet roads, and deployment on edge devices (e.g., Jetson Nano), while focusing on accuracy - speed trade-offs. Finally, it presents future direction such as 3D estimation and self-supervised learning for proactive maintenance.