LMM-Based Indoor Navigation System
via Automatic Map Dictionary Construction

Abstract

In public spaces such as large commercial facilities and train stations, users are often forced to rely on installed signboards and paper maps, making it easy for them to lose track of their current location or misjudge their direction of travel. This research aims to realize an intuitive navigation system to solve this problem. The core of the proposed technology is the application of a Large Multimodal Model (LMM), which possesses both image recognition and language understanding capabilities. By analyzing unstructured map images and converting irregularly placed store names and complex pathway shapes into structured data, the system automatically constructs a map dictionary and a routing graph essential for navigation.