Multimodal Video-Based Breathing Rate Analysis and Applications
Date
2022-01
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
The main goal of this research is to develop a prototype of a video-based system for breathing rate measurement using near-infrared (night-vision) and infrared (thermal) data. This is achieved by identifying patterns that detect abnormalities in the breathing rate, using novel machine learning techniques such as deep neural networks. For the near-infrared spectrum, a pose estimation algorithm is first applied to automatically identify the body areas of interest such as the chest and back regions. Next, these points of interest are tracked over time and used to estimate the breathing rate. The thermal spectrum was also studied in this thesis, in the context of a respiratory disease pandemic. The proposed approach focuses on practical constraints such as wearing face masks which affects the face point tracking. To solve this problem, we propose to use the data from infrared videos of subjects wearing surgical masks in order to detect the mask. The detected mask region is the subject used to pixel intensity analysis which intends to classify the subject’s respiration status as inhaling or exhaling. This is a prerequisite to estimate the breathing rate based on the mask colour variation. The primary application of the proposed approach is to detect healthcare emergencies such as apnea or other breathing abnormalities caused by diseases or induced by drugs. Examples of such applications are provided in this work and reported in the contributed papers. Due to the lack of available datasets with videos of real unhealthy subjects, this thesis investigates the baseline but not real abnormalities, as it analyzes healthy individuals who simulate their breathing.
Description
Keywords
Computer Vision, Deep Learning, Machine Learning, Object Detection, Breathing Rate Estimation, COVID-19
Citation
Pinheiro de Queiroz, L. (2022). Multimodal video-based breathing rate analysis and applications (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.