連續可見光影像的非接觸式呼吸心跳監控系統

Abstract

現代人對自身的健康照護逐漸重視，近年來對生理資訊量測的研究也越來越多，其中呼吸心跳就是生理資訊中最基本但也是最重要的指標，一個人的生理狀況最直接就反映在他的呼吸心跳數值上，所以監控一個人的呼吸心跳資訊就等於是監控一個人的健康與否，但利用傳統接觸式的量測儀器往往會造成許多不便。因此本論文提出基於影像的非接觸式生理資訊量測監控及其應用，利用連續影像測量受測者的呼吸心跳資訊，以解決傳統接觸式量測儀器所造成的問題。 本論文主要分成四大部分：資料前處理、波形處理、獲取生理資訊和資料後處理。第一部分，將一般攝影機獲得的連續影像，利用人臉偵測與改良化的ROI追蹤人臉並減輕晃動的影響，再經過小波轉換(Fast Wavelet Transform, FWT)、Eulerian Magnification(EM)優化影像；第二部分，利用經驗模態分解(Empirical Mode Decomposition, EMD)改善光影變化對偵測的影響；第三部分，利用快速傅立葉轉換(Fast Fourier Transform)和本論文提出的Biorate演算法獲取呼吸心跳資訊；第四部分，利用卡曼濾波器(Kalman Filter)濾除前級獲得的呼吸心跳資訊內所摻雜的雜訊。考慮到運算速度問題，本論文提出合併FWT和EM的加速演算法，以及利用bandpass filter解決EMD疊代多次的問題。

In recent years, there are more and more studies on the physiological monitoring for health care. Breathing rate and heart rate are the basic but important indicators in physiological information. But it caused inconveniences when traditional instruments is used. Therefore, this paper has proposed an image-based physiological monitoring system to measure the human breathing rate and heart rate. In order to solve the measure problems of traditional method. The study contains four parts: data pre-processing, waveform processing, biorate algorithm, and data post-processing. First, capture the frames from a webcam, then use face detection and apply the improved version of ROI to remove the influence of subject’s motion. The original data is enhanced by Fast Wavelet Transform (FWT) and Eulerian Magnification (EM). Second, use an Empirical Mode Decomposition (EMD) to reduce the influence of light. Third, apply Fast Fourier Transform and biorate algorithm to find the breathing rate and heart rate. Fourth, use Kalman filter to filter out the suddenly change of the breathing rate and heart rate. Considering computational load, we combine FWT and EM in one algorithm, and employ a bandpass filter to resolve the iteration problem of EMD.