Learning significant user locations with GPS and GSM

Abstract

This thesis addresses the tasks of place discovery and place recognition - learning and recognizing places significant to a user - by analyzing GPS location and GSM cell tower data collected from the user's mobile phone. Location provides valuable context into the user's environment, and place-discovery and recognition algorithms enable human-centric systems to communicate with the user in human terms. In this thesis, we introduce a novel two-phased approach to place-discovery and recognition that combines the advantages of GPS and GSM cell data. We design and implement a system that produces a compact travel summary from the user's daily GPS logs. We then use computational geometry to investigate the aspect ratios of GSM cell coverage polygons as an optimization to place recognition. Finally, we conclude by presenting a one-month empirical study to demonstrate the effectiveness of our two-phased approach, and identify a set of anomalies in our experiment that can direct further development of place-discovery systems.