Undergraduate Student Scholarship

Permanent URI for this community

https://hdl.handle.net/10323/2095

Browse

Browsing Undergraduate Student Scholarship by Author "Andrews, Jack"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Statistical Learning of Passive Infrared (PIR) Sensor Data for Human Monitoring Applications
    Andrews, Jack; Li, Jia
    Elderly individuals, especially those who suffer from neurodegenerative diseases, are prone to cognitive impairment and deteriorating motor stability. Moreover, with the emerging field of the internet of things (IoT), security of the data shared throughout these connected devices is a growing challenge, leading to the advancement of the field of biometrics. Passive infrared (PIR) sensors are commonly deployed in related monitoring and security applications as they are cheap, non-contact, commercial-off-the-shelf (COTS) components that have proven to be relatively accurate. PIR sensors function by detecting human presence through a change in infrared radiation across the field of the view (FoV) of the sensor. As a result, the major known drawback with PIR sensors is their inability to reliably detect stationary human occupants. Towards expanding the applications and accuracy of PIR sensors, two methods for detecting stationary human subjects are developed and presented: a motion induced PIR sensor (MI-PIR) and a chest motion PIR sensor (CM-PIR). MI-PIR artificially induces the motion necessary for accurate human detection via a robotic actuator, while CM-PIR relies on the movement of the human chest for accurate detection of stationary subjects. The efficacy of these human presence classifications using a single PIR sensor, as well as related human monitoring classifications and regressions, is dependent on statistical learning algorithms for differentiation of scenarios that exist between the numerous applications. A recurrent neural network (RNN) deep learning model with long short-term memory (LSTM) units is found to be superior for PIR sensor data classification due to the time-series nature of the data, and a Gaussian process regression (GPR) machine learning model is utilized as a regression approach for precise indoor localization. The results of this work show that MI-PIR and CM-PIR are both accurate systems for stationary human presence detection and related human monitoring applications using a single PIR sensor.