Geometric, topological & semantic analysis of multi-building floor plan data

Abstract

Generating a comprehensive model of a university campus or other large urban space is a challenging undertaking due to the size, geometric complexity, and levels of rich semantic information contained in inhabited environments. This thesis presents a practical approach to constructing topological models of large environments from labeled floor plan geometry. An exhaustive classification of adjacency types is provided for a university infrastructure including roads, walkways, green-space, and the detailed interior spaces of campus buildings. The system models geospatial features for over 160 buildings within the MIT campus, consisting of more than 800 individual floors, and approximately 36,000 spaces spanning indoor and outdoor terrain. The main motivation is to develop an intuitive, human-centered approach to navigation systems. An application is presented for generating efficient routes between locations on MIT's campus with coverage of both interior and exterior environments. A second application, the MIT WikiMap, aims to generate a more expressive record of the environment by drawing from the knowledge of its inhabitants. The WikiMap provides an interface for collaborative tagging of geographical locations on the MIT campus, designed for interfacing with users to collect semantic data.