GeoAI Approaches for Roadside LiDAR Survey Information Extraction

Road networks are essential public assets, yet their condition has long been assessed through slow, subjective, and hazardous manual inspections. Mobile and terrestrial LiDAR now enable rapid acquisition of dense 3D point clouds of road corridors, but the volume and complexity of these datasets create a new bottleneck: converting raw measurements into actionable information on pavement defects and roadside assets. This review examines recent advances in Geospatial Artificial Intelligence (GeoAI)—combining machine learning, deep learning, and large language models (LLMs) —with a focus on their application to LiDAR-based roadside information extraction and Web-based Geographic Information Systems (WebGIS) dissemination. Analytical GeoAI methods, including point-based neural networks, support direct semantic segmentation of 3D data, though challenges remain around class imbalance, limited annotated datasets, and explainability for safety critical decisions. Generative GeoAI, enabled by LLMs, is increasingly used to automate coding tasks and natural language interaction with spatial databases. To illustrate, we present a proof-of-concept platform that ingests mobile LiDAR, performs heuristic defect and asset extraction, stores results in a spatial database, and exposes them through a 2D/3D WebGIS with natural language querying. The study highlights the potential of GeoAI to streamline infrastructure monitoring workflows and identifies opportunities for future research in scalable segmentation models, interoperable WebGIS frameworks, and trustworthy AI for geospatial decision making.