Maximum likelihood estimation for passive energy-based footstep localization

Abstract

Smart living spaces gather real-time sensor data and use the data to infer, predict, and make decisions. One important way of informing smart living systems is by localizing and tracking occupants. This paper utilizes floor vibration data generated by occupant footsteps—captured by a network of underfloor accelerometers—for passive occupant localization and tracking. A novel maximum likelihood (ML) footstep location estimator is proposed, based on received signal strength/power (RSS) at each sensor location. Localization error variance analysis related to sensor layout (a form of geometric dilution of precision) is studied through deriving and analyzing the theoretical Cramér–Rao lower bound. The proposed localization method does not require knowledge of floor properties, propagation velocity, nor damping. Occupant path tracking is achieved via a Kalman filtering scheme, assuming that an occupant has a zero-mean acceleration. The proposed ML localization method is evaluated using Monte Carlo simulations and using single-occupant walking experiments for 2 different test subjects on a 16 m × 2 m instrumented floor section. Results show superiority of the proposed method to previous RSS footstep localization methods.