An iPPG-Based Device for Pervasive Monitoring of Multi-Dimensional Cardiovascular Hemodynamics

Figures

• 

  Figure 1. (a) Illustration of incident light penetration of different wavelengths through layers of the skin and skin reflection mode. In these wavelength ranges, a longer wavelength represents a higher light penetration rate. Green LED at 530 nm can reach the subpapillary plexus of the dermis layer; (b) Absorption spectra of deoxygenated hemoglobin (Hb), and oxygenated hemoglobin (HbO2). The x-axis represents the wavelength, and the y-axis represents the absorption rate. A local maxima absorption rate at 530 nm (green bar); (c) The iPPG’s AC and DC formation from light intensity rely on the light reflection and the arterial blood flow over time.

• 

  Figure 2. The composition of equipment. Left is the picture of the equipment, including the illumination chamber and a Tablet PC for the user interface (UI). Center picture shows the light-source driver, controller, and camera, all of which have been fixed at the back of the chamber. Right shows the light source’s composition, including the light guide, scattering, and reflective plates. The LED-belt light source is around the rim of the light guide plate. Users can place their chin on the chin support and start video recording and iPPG analysis.

• 

  Figure 3. Experimental setup to record iPPG, PPG and ECG data. Biocapture TM was attached to the left hand’s index finger for PPG signal detection and to the chest for ECG signal detection. Meanwhile, our equipment records facial video with the green light on.

• 

  Figure 4. Original iPPG signal of recording. Multiple biophysiological indices, such as pulse pressure, heart rate variability, breathing rate, perfusion variable information can be derived from the signal. The x-axis is the recorded time frame, and the y-axis is the amplitude. The green line is the change of the original pixel value extracted from the picture. The red line is the breath signal after a 0.2~0.5 Hz bandpass. The blue line was after a 1s window sliding-mean process, and the black line is the smooth signal obtained after passing through a 0.5~5 Hz bandpass filter.

• 

  Figure 5. Facial landmarks detection and ROI extraction. Left two are the original image taken by the camera and the 81 facial features detected. Right two represent the 98 triangular regions formed by combinations of different landmarks and the number of each triangular region.

• 

  Figure 6. (a) Illustration of face region segmentation. The entire face is covered by a set of 98 triangles defined by facial landmarks combinations and each triangle has a dot indicating the location of the center of 80 px × 80px rectangles where the iPPG signal was extracted from; (b) Aligned waveform comparison of signal from meta area 20 and meta area 59; (c) Examples of iPPG signal from different meta regions (19,45,59,89,98,92). X-axis represents the sample points (40 Hz), y-axis represents the intensity of signal.

• 

  Figure 7. Comparison of time difference between each region and reference region. (a) Left face’s data demonstrated; (b) Right face’s data demonstrated. In the upper part of (a,b) is the scatter diagram of the time difference between the six regions (each region is composed of the meta region shown in Tab 1 and each meta-region has 18 sample points.) and the reference region. In the lower part of is the distribution of the time difference between the other five regions and forehead. The vertical axis is the time difference (ms), which is used to show the range of the 95% CI of the mean-of-difference. 5000 samples from BCa bootstrap were taken from the six regions of the left and right faces, and the forehead was taken as reference to calculate the probability distribution of difference between other areas and forehead, and we had a bar with a circle indicated the range of 95% CI of difference mean.

• 

  Figure 8. The phase comparison of the symmetrical regions of the left and right faces; (a) shows lines between the sample points of the corresponding region of the left and right face sides. The horizontal axis is the corresponding region, and the vertical axis is the time difference between the sample point and the reference point. (b) is the distribution map of the difference between the corresponding regions of the left and right faces. The horizontal axis marks the subtracting regions, and the vertical axis represents the time mean-of-difference between regions.

• 

  Figure 9. iPPG imaging analysis. (a) the original image, candidate’s 1of a period the video after EVM process (above row), and the bottom row is the iPPG waveforms of right-cheek (blue) and forehead ( frame image under 530nm’s green light; (b) the Goodness Matrix; (c) image of weighted AC, red means large-amplitude and represents the region with strong pulse-rate intensity; blue means small-amplitude and represents the region with weak pulse-rate information;(d) Time lag between forehead’s iPGG signal and each selected ROI’s iPGG signal.

• 

  Figure 10. Illustration of EVM enhancement and visualized using video frames and iPPG waveform. This figure demonstrates 10 frames of a period the video after EVM process (above row), and the bottom row is the iPPG waveforms of right-cheek (blue) and forehead (red) ROI. Two vertical black line shows the gap of two peaks, cheek reach peak at 31th frame and forehead at 33th as expected.