A Proposal of Model for Heart Rate Estimation while using Wheelchair with 6-axis Sensor

Abstract

Recently, the spread of wearable devices makes the measuring and analyzing of various sensors data easy. In the field of sports, studies using human behavior data measured by these devices are actively conducted for the purpose of assisting the human exercise. Most studies, however, are for healthy subjects and few for disabled subjects. The purpose of this study was to suggest the model for the heart rate estimation in driving wheelchair using 6-axis sensors (accelerations and angular velocities) and to assist the exercise of wheelchair users. The suggested model estimated the heart rate from data of 6-axis sensors (attached to under the seat of wheelchair) using the machine learning. The suggested model consists of LSTM and fully-connected. Input to the suggested model were the driving direction acceleration, slope of road and oxygen intake. The driving direction acceleration was calculated by the half-wave rectifying the acceleration of the driving direction axis. The slope was calculated by madgwick filter. The oxygen intake was calculated by the formula derived in the previous experiment. 6-axis sensors data were smoothed by half-overlap with 24 seconds window. The suggested model estimated the heart rate every 12 seconds. When the suggested model was applied to the heart rate estimation in driving wheelchair, it was confirmed to be possible of estimation with 9.31 bpm mean absolute error. Furthermore, it was suggested that the estimation accuracy could be improved by the advanced calculation of the oxygen intake and introducing the parameters of the individual physical ability levels.