Thermal Image based Remote Heart Rate Measurement on Dynamic Subjects Using Deep Learning

In this work, we propose a method of heart rate estimation by using the infrared thermal images as input. When the heart beats that causes the blood vessels to periodically vasoconstriction and vasodilation, there will be a tiny periodic temperature change on face area. Therefore, the temporal signal from thermal images is considered as the input to our designed DEWNet (DEnseWave Net) CNN model for heart rate estimation. In the experiment, we collected a dataset containing 24 subjects who were running on the treadmill with five different running speeds. Consequently, using this dataset the average error is 13.592 bpm, which shows its feasibility for real-world applications.