Identification of building floors in a 3D city model

This paper presents a methodology to identify building floors within a 3D city model using a large data set of Wi-Fi positioning and barometer data collected by about 50,000 students in Singapore. We defined discrete gaps in air pressure clusters in 0.0001 lat, lon grid cells at 60 second intervals. Clusters of pressure points are indicative of individual floors in buildings. Using this method, we find that 1% of SG's population can cover ~5% of all built-up area in the city. We also constructed a citywide 3D path-search engine by applying A*-search to crowd densities rather than shortest distances. This method doesn't require the use of any a priori information such as floor plans, and is computationally efficient.