Indoor navigation for passengers in underground transit stations using smartphones

Abstract

People are increasingly relying on smartphones to solve a lot of their daily problems, among which navigation is one of the most fundamental tasks. Estimating the locations of pedestrians and tracking them in an indoor environment is a long sought after research goal. This thesis develops a smartphone-based indoor navigation system specifically designed for transit stations, but which also allows easy extension to other navigation scenarios. The thesis' system requires no extra hardware installation in the indoor environment or on the phone and few assumptions are made about the indoor space and the path tile user is taking, unlike previous approaches. A Bayesian feature-based particle filter localization model is developed to estimate the user's location. A motion model with step detection and heading inference is developed from phone sensor readings, which serves as the motion input to the particle filter. The thesis develops several human activity pattern recognition models that extract activity features from phone sensors as the observation model in the feature-based particle filter model. A grid-based map representation is developed to model the topology and semantic information of an indoor environment, which requires lower computational cost in real-time particle propagation than 2D geometric maps. The thesis develops a modified shortest path algorithm that is able to accommodate user-specific routing requirements and constraints, such as handicap accessibility and a sequence of locations to be visited. A routing graph that is able to model different types of locations and connections in the indoor environment is also developed to work with the modified shortest path algorithm. A new data model and standardized data collection process are proposed to improve data quality and the user experience in future indoor path planning applications. Last, an integrated indoor navigation system is developed to provide the user with step-by-step instructions and route display. Empirical studies of system performance are performed for several transit stations in Boston and London, and a set of buildings at MIT.