Contact-free motion-robust vital signs monitoring

• Main Author: Zheng, Tianyue
• Other Authors: Luo Jun
• Format: Thesis-Doctor of Philosophy
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/170017
• Institution: Nanyang Technological University

Vital signs monitoring plays a crucial role in healthcare and biomedical applications, as it enables healthcare professionals to assess physiological status and make informed clinical decisions. Conventional methods for vital signs monitoring typically involve wearable sensors directly attached to the body. However, these methods are often not well received by human subjects due to the discomfort and inconvenience associated with long-term usage. Consequently, researchers have been pursuing contact-free alternatives to address these limitations. Despite ongoing efforts, most existing contact-free designs require human subjects to remain still, thus limiting their practicality in everyday environments where motion-induced interference is inevitable. This thesis aims to investigate the effects of motion-induced interference on contact-free vital signs monitoring, and proposes systems that use signal processing as well as deep learning algorithms to deliver motion-robust solutions. The proposed systems remain effective even in scenarios with motion-induced interference, such as monitoring in moving vehicles or during body movements, making them suitable for use in diverse everyday environments. The first scenario of interest in this thesis involves monitoring vital signs in moving vehicles. Given the significant amount of time people spend in vehicles, health issues under driving conditions have become a major concern. Such issues may vary from fatigue, asthma, stroke, to even heart attack, yet they can be adequately indicated by various vital signs. Consequently, in-vehicle vital sign monitoring can help us predict and hence prevent these issues. Whereas existing wearable sensor- and camera-based methods could detect these indicators, intrusiveness, privacy concerns, and system complexity all call for a convenient yet effective and robust alternative. This thesis first aims to develop V2iFi, an intelligent system performing monitoring tasks using a commercial-grade IR-UWB radar mounted on the windshield. With the help of an algorithm based on signal processing, V2iFi is capable of reliably detecting driver's vital signs under driving conditions and in the presence of passengers, thus allowing for potentially inferring corresponding health issues. Despite its capability to remove motion-induced interference from moving vehicles, V2iFi falls short when it comes to handling stronger interference caused by body movements, motivating the exploration of new approaches throughout the remainder of this thesis. To improve the vital signs monitoring under full-scale body movements, this thesis proposes MoRe-Fi, which enables fine-grained monitoring of respiratory waveforms through the use of an IR-UWB radar and deep learning. MoRe-Fi fully exploits the complex radar signal to perform data augmentation. The core of MoRe-Fi is a novel variational encoder-decoder deep-learning network that reconstructs respiratory waveforms from noise and interference. In comparison to the signal processing techniques adopted by V2iFi, the deep learning algorithm is more powerful and is able to isolate respiratory waveforms modulated by body movements in a non-linear manner. However, MoRe-Fi still cannot handle human subjects during walking, which is a common scenario in daily life. This motivates us to further explore this scenario in the rest of this thesis. To overcome the aforementioned limitation, the thesis also proposes BreathCatcher, which achieves simultaneous tracking and respiration monitoring, thus enabling vital signs monitoring for even walking human subjects. To deal with the even stronger motion-induced interference during walking, BreathCatcher enhances the variational encoder-decoder network by using temporal convolutional layers capturing long-range dependencies in data samples. It also utilizes a hybrid human respiration and position-tracking algorithm to locate and identify respiratory signals from complex RF reflection mixtures. Essentially, BreathCatcher can obtain respiratory waveforms from multiple walking human subjects and identify each subject according to the latent properties of the respiratory signals. In summary, the proposed systems offer convenient, effective, and robust alternatives to conventional wearable sensor-based methods for monitoring vital signs. Moreover, they cater to diverse application scenarios, including in-vehicle monitoring and indoor monitoring under full body movements and even walking. Our experiments evidently demonstrate the accuracy of vital signs monitoring using the proposed systems in diverse scenarios. The potential applications of the proposed system for healthcare and biomedical applications are significant, particularly in the context of predicting and preventing health issues under challenging conditions.