Abstract

Contactless detection of human vital sign using radar sensors appears to be a promising technology which integrates communication, biomedicine, computer science etc. The radar-based vital sign detection has been actively investigated in the past decade. Due to the advantages such as wide bandwidth, high resolution, small and portable size etc., ultra-wideband (UWB) radar has received a great deal of attention in the health care field. In this thesis, an X4 series UWB radar developed by Xethru Company is adopted to detect human breathing signals through the radar echo reflected by the chest wall movement caused by breath and heartbeat. The emphasis is placed on the estimation of breathing and heart rate based on several signal processing algorithms.

Firstly, the research trend of vital sign detection using radar technology is reviewed, based on which the advantages of contactless detection and UWB radar-based technology are highlighted. Then theoretical basis and core algorithms of radar signals detection are presented. Meanwhile, the detection system based on Xethru UWB radar is introduced. Next, several preprocessing methods including SVD-based clutter and noise removal algorithms, the largest variance-based target detection method, and the autocorrelation-based breathing-like signal identification method are investigated, to extract the significant component containing the vital signs from the received raw radar echo signal. Then the thesis investigates four time-frequency analysis algorithms (fast Fourier transform + band-pass filter (FFT+BPF), empirical mode decomposition (EMD), ensemble empirical mode decomposition (EEMD) and variational mode decomposition (VMD) and compare their performances in estimating breathing rate (BR) and heart rate (HR) in different application scenarios.

A python-based vital signs detection system is designed to implement the above-mentioned preprocessing and BR and HR estimation algorithms, based on which a large number of single target experiments are undertaken to evaluate the four estimation algorithms. Specifically, the single target experiments are divided into simple setup and challenging setup. In the simple setup, testees face to radar and keep normal breathing in an almost stationary posture, while in the challenging setup, the testee is allowed to do more actions, such as site sitting, changing the breathing frequency, deep hold the breathing. It is shown that the FFT+BPF algorithm gives the highest accuracy and the fastest calculation speed under the simple setup, while in a challenging setup, the VMD algorithm has the highest accuracy and the widest applicability.

Finally, double targets breathing signal detection at different distances to the radar are undertaken, aiming to observe whether the breathing signals of two targets will interfere with each other. We found that when two objects are not located at the same distance to the radar, the object closer to the radar will not affect the breath detection of the object far from the radar. When the two targets are located at the same distance, the 'Shading effect' appears in the two breathing signals and only VMD algorithm can separate the breathing signals of the targets.