Heartbeat detection with Doppler sensor using adaptive scale factor selection on learning

Heart rate variability gives information about health and mental condition. Noncontact detection of heartbeat using Doppler sensor has been researched in many studies. There is a major issue which is how to reduce the influence of body movement. A conventional algorithm uses the continuous wavelet transform. To extract heartbeat, a constant scale factor is selected during a learning phase which is then used to detect heartbeat during a test phase. However, to select the scale factor, the authors do not consider the difference of heart rate between learning and test. Thus, the root mean square error (RMSE) of R-R interval which represents the peak-to-peak of heartbeat is deteriorated. In this paper, we propose a method to improve the RMSE of R-R interval compared with the conventional one. During learning, we search for a scale factor interval corresponding to the heart rate obtained with the Doppler sensor. To take the difference of heart rate between learning and test into consideration, we extend the scale factor interval depending on the action during test. After we select a certain scale factor from some scale factors in the extended interval, we detect heart rate during test by counting the peaks of wavelet coefficients of the selected scale factor. Through experiments, when a subject is sitting still or doing a typing game, we show that the RMSE of R-R interval is improved by about 60 msec and 65 msec, respectively, compared with the conventional method.