Deep learning is already showing good results in roadway asset classification, which plays a key role in safety management, regulatory compliance, and asset inventory in Transportation Asset Management (TAM). You Only Look Once (YOLO), Single Shot MultiBox Detector (SSD), and Faster R-CNN are models that are credible at detecting roadside structures, lane divisions, and traffic signs in varied conditions. In addition to approaches based on images, three-dimensional mapping (Light Detection and Ranging or LiDAR) makes classification processes more accurate, as there are geometric depth features that improve the ability to distinguish between visually similar objects, especially in high-density environments. Other, more recent, multimodal fusion approaches (LiDAR point clouds, RGB images, multispectral satellite images, and street-view images) enhance robustness in scenes where either there is occlusion or adverse lighting conditions. Taken together, these developments are compatible with scalable, automated roadway inventories, asset inventories that are compatible with TAM goals.
Keywords: automated roadway inspection, deep learning for infrastructure monitoring, high-resolution satellite imagery, pavement condition index (PCI) prediction, roadway asset classification
