Breathing Pattern Monitoring using Remote Sensors

TL;DR

This paper presents a novel, non-contact method for monitoring breathing patterns using thermal and RGB cameras, achieving high classification accuracy for various respiratory conditions, enabling unobtrusive health assessment.

Contribution

Introduces a new algorithm for extracting respiratory features from thermal and RGB camera data and demonstrates its effectiveness in classifying respiratory patterns.

Findings

Achieved up to 95.79% classification accuracy.

Compared multiple respiratory signal extraction algorithms.

Validated the approach for diverse respiratory patterns.

Abstract

Breathing is one of the most important body functions because it provides it with oxygen, which is vital for energy production. In addition, the removal of carbon dioxide actively regulates the acid-base level, which is essential for the physiological function of the body. Due to its close connection with many other body functions, respiration can also be used as an indicator for a wide spectrum of medical conditions, which at first glance have little to do with breathing. Neurological, cardiological, inflammatory, metabolic, and even psychological conditions symptomatically show up in breathing patterns. Hence, being able to classify them automatically and unobtrusively, can allow cost-effective monitoring systems to continuously assess the health of a patient. In this work, multiple respiratory signal-extraction algorithms for thermal and RGB cameras are presented and compared. A novel algorithm for the extraction of multiple respiratory features is presented and evaluated. Using a one vs. one multiclass support vector machine, these features were used to classify a wide range of respiratory patterns with an accuracy of up to 95.79%.