Short-time Estimation of R-R Interval from Facial Video Image with a Multiple-Measurement-Points-Voting-Method
Kazutoshi Ukai, Rashedur Rahman, Syoji Kobashi
Modern society is called a stressful society due to long working hours with electronic equipment and/or human relationships in workplaces, and hence periodic stress checking is required. It is known that there is a relationship between heart rate variability and mental stress. Detection of R-R Interval (RRI) from facial video images is expected to enable noncontact and unconsciously monitoring of mental stress. This study proposes a high-accuracy and short-time RRI estimation method in facial video images, which is called multiple-measurement-points-voting-method, MMVM in short. MMVM can detect RRI of individual facial pixels with time-course color signal change,and estimate RRI from the detected RRIs of multiple points. MMVM was applied to 252 trials of 32 subjects, and was compared with ground truth RRI measured using plethysmographic sensor. The mean absolute error of MMVM was 17.7 ms. And, the minimum measurement duration to reach target accuracy of MMVM was 2.8 sec, and that of ICA method was 5.3 sec. MMVM shortens the measurement time by about 47%.