Human vital signs are crucial parameters which reflect essential body functions and are often accessed by medical professionals at the first place during clinical diagnostics to provide immediate assistance in health status measurements. However, due to the recent COVID-19 pandemic, measurements made with direct body contact have become increasingly challenging and costly because of the spreading nature of this virus. Therefore, contactless vital sign measurements are highly desirable, and it motivates us to research and develop a new solution which is capable of performing real time heart rate (HR) detection, respiratory (RR) detection, and body temperature (BT) measurement together from a distant human subject under an ambient light environment. The thesis describes a new system framework, which utilizes the power of computer vision to collect remote video image data, processes them using signal processing and machine learning (ML) technologies simultaneously, and produces rapid updates on display. Furthermore, our validation analysis on the system has showed varied results based on different methodologies used, which enables us to apply the most suitable approach on each component for an optimized final integration.
At the time of completing this thesis, we have achieved a complete system integrated with remote HR, RR estimations and BT detection, which are all fully functional in both real-time and offline. To further refine the performance on HR estimation, we selected the extreme gradient boost model through a number of ML models we tested, as it not only gives the lowest root mean square error of 8.2 but also produces stable and robust output.
Identifer | oai:union.ndltd.org:uottawa.ca/oai:ruor.uottawa.ca:10393/41111 |
Date | 28 September 2020 |
Creators | Ma, Xiaocong |
Contributors | El Saddik, Abdulmotaleb |
Publisher | Université d'Ottawa / University of Ottawa |
Source Sets | Université d’Ottawa |
Language | English |
Detected Language | English |
Type | Thesis |
Format | application/pdf |
Page generated in 0.0017 seconds