In this thesis a proof-of-concept of passive indoor localization system using visible light is proposed that does not require active participation of a user in the localization process. The user neither holds any device nor do they have any sensor tags attached to their body. The system can be implemented by employing existing lighting infrastructure that is used in visible light communication systems. The sources and receivers can be arranged in any form in the room, but in this work they are considered co-located on the ceiling. This arrangement is advantageous since it reduces installation complexity and is most commonly used in indoor environments. The proposed approach measures impulse response (IR) between the source-receiver pairs in order to localize a localization object (LO) i.e., the user. The presence of the LO inside the room alters the IR measurements between the source-receiver pairs that can be related to its position. The changes in measured IRs are leveraged for position estimation. The proposed research work can be divided into two main parts as follows. In the first part, a single-bounce reflection model of light rays is considered and the room contains only LO inside it. The fingerprinting method is used to estimate position of the LO and analytical expression of Cramer-Rao lower bound is derived on the positioning error. In the second part, a realistic room model using reasonable parameters and multi-order reflections is considered where the furniture is also placed inside the room. A deep learning framework is employed that learns changes in IRs corresponding to random locations of LO in the room in order to estimate its position from the unknown IR measurements. Furthermore, a fall detection system is developed that classifies upright or prone states of the LO from single set of IR measurements.

Doctor of Philosophy (PhD)

Thesis