Adding identity to device-free localization systems

Abstract

Recent advances in wireless localization systems show that by transmitting a wireless signal and analyzing its reflections, one can localize a person and track her vital signs without any wearables. These systems can localize with high accuracy even when multiple people are present in the environment. However, a primary limitation is that they cannot identify people and know who is the monitored person. In this thesis, we present a system for identifying people based only on their wireless reflection with high accuracy. We use a semi-supervised learning classifier to assign labels to each person tracked by the device-free localization system. We use recent advances in machine learning to leverage the big amount of unsupervised data that we have. A key challenge that we solve is obtaining labels that are used for guiding the classifier. To get labeled data, we devised a novel scheme to combine data from a sensor that people are carrying with data from a wireless localization system. We deployed and evaluated our system in people's homes. We present a case study of how it can be helpful to monitor people's health more effectively.