AURES

Abstract

In this paper, we present a platform designed for low-power real-time sensing of the number of occupants in indoor spaces. The system transmits a wide-band ultrasonic signal into a room and then processes the superposition of the reflections recorded by a microphone. The system has two modes of operation, one for presence detection and one for estimating the number of occupants in a region. The presence detection uses the difference between multiple transmissions in succession with a set of general classifiers that make a binary decision about if the room contains occupants. We then use a semi-supervised learning approach based on Weighted Principal Component Analysis (WPCA) that requires minimal training data to estimate the number of occupants. We also present the design of an energy harvesting embedded platform and demonstrate that our algorithm can continuously execute using energy harvested from indoor solar panels. The platform has a dual Bluetooth Low-Energy and 802.15.4 interface to communicate with a gateway or nearby mobile phone that runs an interface that aids in collecting training data. We evaluate our algorithm on a wide-variety of indoor spaces as well as benchmark the hardware in terms of sampling rate given an energy budget. On more than three weeks of data, we see that we can detect motions with an average of 85% recall rate and perform occupancy counting with an average error of 10% in terms of maximum occupancy.