Data Processing for Device-Free Fine-Grained Occupancy Sensing Using Infrared Sensors

Abstract

Fine-grained occupancy information plays an essential role for various emerging applications in smart homes, such as personalized thermal comfort control and human behavior analysis. Existing occupancy sensors, such as passive infrared (PIR) sensors generally provide limited coarse information such as motion. However, the detection of fine-grained occupancy information such as stationary presence, posture, identification, and activity tracking can be enabled with the advance of sensor technologies. Among these, infrared sensing is a low-cost, device-free, and privacy-preserving choice that detects the fluctuation (PIR sensors) or the thermal profiles (thermopile array sensors) from objects' infrared radiation. This work focuses on developing data processing models towards fine-grained occupancy sensing using the synchronized low-energy electronically chopped PIR (SLEEPIR) sensor or the thermopile array sensors.

The main contributions of this dissertation include: (1) creating and validating the mathematical model of the SLEEPIR sensor output towards stationary occupancy detection; (2) developing the SLEEPIR detection algorithm using statistical features and long-short term memory (LSTM) deep learning; (3) building machine learning framework for posture detection and activity tracking using thermopile array sensors; and (4) creating convolutional neural network (CNN) models for facing direction detection and identification using thermopile array sensors.