Heart Rate Estimation from Facial Videos Based on Convolutional Neural Network

摘要

Recent research about remote heart rate (pulse) estimation through facial videos has aimed to improve the measurement accuracy under noisy interference by improving the preprocessing method to get more robust time domain feature representation. This paper presents a pulse measurement method based on deep learning framework. Our method uses Convolutional Neural Network (CNN) as a regression model on the spectrum images, which combined processed time domain features with its respective frequency feature. The experimental results show that the effect of our estimation method can overcome the interference caused by light variation and head motions, and it obtains superior performance compare to the most previous works.