Smartphone Localization for Indoor Pedestrian Navigation

Abstract

Global positioning system (GPS) interfacing with applications such as Google Maps has proven very useful for navigation in outdoor open settings. However in crowded metropolitan environments with high rise buildings or in indoor settings, GPS quickly becomes unreliable. Using sensors found on commodity smartphones to perform accurate pedestrian localization in complex indoor settings remains a challenging problem. RSSI based methods provide absolute positioning but provide sparse updates and require a dense network of beacons. Inertial Measurement Unit (IMU) based methods provide high rate, smooth, relative positioning but are very sensitive to noise and drift with time. In this thesis we propose a smartphone based indoor localization system which combines the absolute positioning from Bluetooth RSSI based localization with the smooth relative positioning from IMU based inertial odometry and map information to fix drift. This system reduces the number of beacons required for accurate localization by a factor of 5 compared to state- of-the-art RSSI only methods. Compared to state-of-the-art IMU only methods, our method is 2.4x more accurate and compared to state-of-the- art inertial map-prior networks it is 19% more accurate. We deployed the deep models to mobile phones in the form of iOS and Android apps to run real-time localization thereby facilitating path planning and turn-by-turn navigation.