Device-Free Localization Using Privacy-Preserving Infrared Signatures Acquired From Thermopiles and Machine Learning

Abstract

The development of an accurate passive localization system utilizing thermopile sensing and artificial intelligence is discussed in this paper. Several machine learning techniques are explored to create robust angular and radius coordinate models for a localization target with respect to thermopile sensors. These models are leveraged to develop a reconfigurable passive localization system that can use a varying number of thermopiles without the need for retraining. The proposed robust system achieves high localization accuracy (with the median error between 0.13 m and 0.2 m) while being trained using a single human subject and tested against multiple other subjects. It is shown that the proposed system does not experience any significant performance deterioration when localizing a subject at different ambient temperatures or with different configurations of the thermopile sensors placement.

Keywords

• Device-free localization (DFL)
• human sensing
• indoor positioning system (IPS)
• infrared sensing
• machine learning
• passive localization