Occupant Monitoring using Footstep-Induced Floor Vibration
2020-01-24T16:06:11Z (GMT) by
The overall objective of this research is to monitor occupants in indoor settings using their footstep-induced floor vibration. Some of the current sensing approaches for occupant monitoring include vision-based, radio-frequency-based, pressure-based, and mobile-based sensing. However, maintenance and installment requirements, such as dense deployment and requiring the occupants to carry a device, limit their application. To overcome these limitations, we have introduced vibration-based sensing as a sparse and non-intrusive alternative for occupant monitoring which does not require carrying a device. The intuition behind this sensing approach is that occupant footsteps cause floor vibration waves which travel through the structure and reach our sensors. These vibrations can be used for extracting information about the occupants (e.g., location and presence). However, because these vibration waves travel through the structure, they are also affected by the structural characteristics which result in various research challenges for occupant monitoring. In this dissertation, I have focused on three research contributions which are based on structure-related challenges. First, we present a floor-vibration-based occupant detection approach which enables detection across different structures through “model transfer”. The structural effects on the signals results in footstep models being different in different structures which consequently adds to the labeled data and calibration requirements. To address this challenge, we characterize the effect of the structure on the footstep-induced floor vibration responses to develop a physics-driven model transfer approach that enables step-level occupant detection across structures. Specifically, our model transfer approach projects the data into a feature space in which the structural effects are minimized. By minimizing the structure effect in this projected feature space, the footstep models mainly represent the differences in the excitation types and therefore are transferable across structures. In other words, in this projected space, a footstep model trained in a source structure in which labeled data is available can be used for target structures in which no labeled data is available. By only requiring labelled data from a few source structures, this approach significantly reduces the labeled data and calibration requirements. We analytically show that the structural effects are correlated to the Maximum-Mean-Discrepancy (MMD) distance between the source and target marginal data distributions. Therefore, to reduce the structural effect, we minimize the MMD between the distributions in the source and target structures. We evaluated the robustness of our approach through field experiments in three types of structures. Our evaluation consists of training a footstep model in a set of structures and testing it in a different structure. As the performance metric, we have utilized F1 score which is the harmonic mean of the precision and recall rate and has been commonly used for evaluating classification algorithms. Across the three structures, the evaluation results show footstep detection F1-score of up to 99 percent for our approach, corresponding to 29X improvement compared to a baseline approach which does not transfer the model. Second, we characterize dispersive wave propagation to localize occupants using their footstep-induced floor vibrations and without extensive calibration. To localize the footsteps, we utilize the Time Differences of Arrival (TDoA) and the propagation velocity of the footstep-induced vibration waves. To this end, the main challenges are: 1) the vibration wave propagation in the floor is of dispersive nature (i.e., different frequency components travel at different velocities) and 2) due to floor heterogeneity, these wave propagation velocities vary in different structures as well as in different locations in a structure. These challenges result in signal distortions which in turn reduce the TDoA and propagation velocity estimation accuracy and lead to large localization inaccuracies or calibration requirements. We present a “decomposition-based dispersion mitigation technique” which extracts specific components (which have similar propagation characteristics) for localization. Further, we introduce an “adaptive multilateration approach” that employs heuristics to constrain the search space and robustly locate the footsteps when the propagation velocity is unknown. We evaluated our approach using field experiments in 3 different types of buildings (both commercial and residential) with human participants. The results show an average localization error of 0.34 meters, which corresponds to a 6X reduction in error compared to a baseline method (which will be defined in the thesis). Furthermore, our approach resulted in standard deviation of as low as 0.18 meters, which corresponds to a 8.7X improvement in precision compared to the baseline approach. Third, we model the obstruction effect on the footstep-induced floor vibration waves to enable robust occupant localization in obstructive indoor settings. obstructions such as walls and furniture add mass to the structure which affect the structural characteristics, the wave propagation velocity, and in turn, reduce the localization performance. To address this challenge, we localize footsteps by considering different velocities between the footsteps and sensors depending on the existence and mass of obstruction on the wave path. Specifically, we 1) detect and estimate the mass of the obstruction by characterizing the wave attenuation rate, and 2) use this estimated mass to find the propagation velocities for localization by modeling the velocity-mass relationship through the lamb wave characteristics. Finally, we leverage these propagation velocities to locate the footsteps (and the occupants) using our non-isotropic multialteration approach. In field experiments, we achieved average localization error of 0.61 meters, which is 1) the same as the average localization error when there is no obstruction and 2) 1.6X improvement compared to a baseline approach which does not consider the effect of the obstruction.