Heart Rate Estimation by Near-Infrared-iPPG Considering Motion Artifacts During Sleep

In recent years, the number of people complaining about inappropriate sleeping habits has been increasing annually, and non-contact measurement of vital signs during sleep has attracted attention. A non-contact heart rate measurement method using an infrared camera was developed. However, camera-based heart rate estimation is problematic because motion artifacts significantly impact measurement results. The conventional method combines multiple heart rate estimation methods or makes appropriate peak selections using the coordinates of the facial image to perform heart rate estimation that considers motion artifacts. However, motion artifacts are caused not only by changes in facial position but also by changes in angle. Conventional methods cannot properly remove body motion, which affects the heart rate estimation system. Therefore, this study proposes motion artifact removal by combining an adaptive filter with singular spectrum analysis. Additionally, the Euler angle information of the head extracted using a convolutional neural network was incorporated as a parameter to estimate the heart rate corresponding to the motion artifacts caused by angular displacement. The estimation accuracy was demonstrated through the experimental results.