[2H1-OS-28] OS-28

9:15 AM - 9:30 AM JST(12:15 AM - 12:30 AM UTC)

[2H1-OS-28-02] GeoAI-powered Route Generation from Map Photos Using Multimodal LLMs

〇Hanqi Li¹, Kei Hiroi¹, Michinori Hatayama¹ (1. Kyoto University)

Keywords:

AI,GIS,Computer Vision

This paper presents a GeoAI system that automatically generates routes on map images based on natural language user requests. When a user uploads a photograph of a physical map and inputs a query (e.g., "Show me the shortest path to the hospital"), the system interprets both the visual map content and the textual request to draw an appropriate route directly on the map image. Our approach leverages multimodal Large Language Models (LLMs) to understand user intent and recognize road networks from map photographs. The recognized spatial information is combined with computer vision and GIS processing methods to achieve accurate route calculation and visualization. Experimental results demonstrate that our system successfully generates correct routes on real map photographs in response to various user queries.