Heart Rate Estimation during Exercise from Photoplethysmographic Signals Using Convolutional Neural Network

Abstract

Heart rate (HR) is a useful indicator for making physical exercises efficient as HR reflects activity load. One of the most convenient methods for measuring HR is to use on-wrist photoplethysmographic (PPG) signals while its estimation accuracy may be deteriorated when there exist large motion artifacts due to exercise. In this paper, a novel method for HR estimation using convolutional neural network (CNN) is proposed. The augmented encoding suitable for HR estimation is devised for improving estimation accuracy and training efficiency. With the use of CNN, features unique to each person can be acquired without giving handcrafted features. The experimental results show that the average absolute error of the proposed method is 4.12 BPM whereas that of FFT is 5.00 BPM.