Recovering pulse rate during motion artifact with a multi-imager array for non-contact imaging photoplethysmography

Photoplethysmography relies on characteristic changes in the optical absorption of tissue due to pulsatile (arterial) blood flow in peripheral vasculature. Sensors for observing the photoplethysmographic effect have traditionally required contact with the skin surface. Recent advances in non-contact imaging photoplethysmography have demonstrated that measures of cardiopulmonary system state, such as pulse rate, pulse rate variability, and respiration rate, can be obtained from a participant by imaging their face under relatively motionless conditions. A critical limitation in this method that must be resolved is the inability to recover these measures under conditions of head motion artifact. To investigate the adequacy of channel space dimensionality for the use of blind source separation in this context, nine synchronized, visible spectrum imagers positioned in a semicircular array centered on the imaged participant were used for data acquisition in a controlled lighting environment. Three-lead electrocardiogram and finger-tip reflectance photoplethysmogram were also recorded as ground truth signals. Controlled head motion artifact trial conditions were compared to trials in which the participant remained stationary, with and without the aid of a chinrest. Bootstrapped means of one-minute, non-overlapping trial segments show that, for situations involving little to no head motion, a single imager is sufficient for recovering pulse rate with an average absolute error of less than two beats per minute. However, error in the recovered pulse rate measurement for the single imager can be as high as twenty-two beats per minute when head motion artifact is severe. This increase in measurement error during motion artifact was mitigated by increasing the dimensionality of the imager channel space with multiple imagers in the array prior to applying blind source separation. In contrast to single-imager results, the multi-imager channel space resulted in an absolute error in the recovered pulse rate measurement that is comparable with pulse rate measured via fingertip reflectance photoplethysmography. These results demonstrate that non-contact, imaging photoplethysmography can be accurate in the presence of head motion artifact when a multi-imager array is implemented to increase the dimensionality of the decomposed channel space.