Research on the implementation of road inspections in municipalities using 3D point cloud data via MMS and Large Lan-guage Models (LLMs)

Abstract

Research on using new technologies and AI for road inspections has advanced in recent years. Many municipalities want to implement new technologies to improve operational efficiency. However, the specific benefits and challenges are unclear, so implementation has not occurred. This paper examines the effects and issues of road pavement and lane marking inspections. It uses 3D point cloud data from MMS (Mobile Mapping System) surveys and the Large Language Model(LLM) Gemini. Pave-mentinspection used the cracking ratio, rut depth, and IRI (International Roughness Index). Lane marking inspection used a five-level visual inspection ranking for outer and center lines to assess soundness. LLM inspection accuracy was high for cracking ratio. Human inspection using 3D point cloud data from MMS surveys showed high accuracy for cracking ratio, rut depth, and lane marking soundness.