Multi-ROI Spectral Approach for the Continuous Remote Cardio-Respiratory Monitoring from Mobile Device Built-In Cameras

Figures

• 

  Figure 1. Flowchart presenting the steps carried out to extract the breathing waveform from the ROI torso for continuous estimation of respiratory frequency.

• 

  Figure 2. Flowchart presents the steps carried out to extract the r-PPG signals from image sequences for continuous heart rate estimation.

• 

  Figure 3. (A) Schematic representation of the experimental setup. (B) Picture of a subject during the recording of the video in the different experimental conditions (see Table 1 for the operating conditions).

• 

  Figure 4. Example of raw r-PPG signals extracted from the three ROIs and the breathing signal extracted by the OF technique from the ROI torso.

• 

  Figure 5. Temporal trends of estimated fR and HR values against the reference values.

• 

  Figure 6. Example of HR values obtained from the windowing analysis. In black, the reference HR values, in green the outlier values; in blue, the estimated HR values after the outlier removal and their replacement.

• 

  Figure 7. Bland–Altman plots comparing fR estimated through optical flow in different ambient conditions: (A) when the light ring is on; (B) when there is only natural light; (C) at a distance of 0.5 m; (D) at a distance of 1.0 m.

• 

  Figure 8. Correlation analysis between the HR values estimated with the implemented post-processing techniques and the reference system in different conditions: (A) when the light ring is on; (B) when there is only natural light; (C) at a distance of 0.5 m; (D) at a distance of 1.0 m. The darkest color represents the minimum value of R2, whereas the lightest color is the highest value of R2.

• 

  Figure 9. Bland–Altman plots comparing the average HR estimated by using six different post-processing techniques in different ambient conditions.

• 

  Figure 10. Bland–Altman plots comparing HR estimated from the r-PPG signals with the SNR-based approach.