Device-Free Localization Using Low-Resolution Thermopiles

Abstract

Low-resolution thermopile sensors have emerged as a promising solution for privacy-preserving indoor localization. In this study, we investigate the localization performance of custom-designed thermopile sensors. We systematically evaluate multiple ceiling- and wall-mounted configurations, addressing the limitations of prior works that examined only a small number of sensors and fixed layout. A 2D CNN–LSTM regression model was trained and evaluated with data collected from 8 participants achieving median localization errors between 0.17–0.22 m. Our findings indicate a clear dependence of localization accuracy on the number and arrangement of sensors. Experimental benchmarking against previous approaches indicates that the proposed algorithm provides improved localization accuracy. To assess scalability, we deployed four ceiling-mounted sensors to cover a substantially larger area than the state-of-the art and demonstrated that our proposed approach perform effectively in such a setting. All hardware design information and code are made publicly available to support continued research.