Video heart rate measurements based on adaptive region of interest

Video-based non-contact photoplethysmography (PPG) can achieve non-contact heart rate monitoring. To improve the quality of non-contact PPG signal and the accuracy of non-contact heart rate, a novel method based on adaptive region of interest (ROI) was studied. First, independent vector analysis (IVA) was used to process the face by blocks. Then normalized cuts (Ncuts) was used to select blocks with the highest signal-to-noise ratio and correlation as adaptive ROI to obtain the heart rate. Finally, the non-contact PPG signal was obtained by weighted average of the adaptive ROI and frequency domain processing. Compared with pre-selecting ROI method, the mean and standard deviation of the heart rate error was reduced from (4.72±6.46) beats per minute (bpm) to (0.52±1.49) bpm, the root mean square error (RMSE) was reduced from 7.96bpm to 1.50bpm, and the average error rate was reduced from 9.45% to 1.73% in static situation. The error of this method under head movement was (1.02±2.91) bpm, the RMSE was 2.11 bpm, which was reduced by above 50%. Bland-Altman and correlation analysis was used to compare the heart rate obtained by this method and using a contact PPG instrument, the 95% confidence interval was -2.44~3.48 bpm at static and -2.76~4.79 bpm at move situation. Compared with the waveform of the contact PPG signal, it was showed the method obtains the PPG signal with complete details. Experimental results showed this method has significant advantages in the quality of PPG signal and the accuracy of heart rate. © 2021, Science Press. All right reserved.