Abstract

Heart rate is an important indicator of people’s physio-logical state. Recently, several papers reported methods tomeasure heart rate remotely from face videos. Those meth-ods work well on stationary subjects under well controlledconditions, but their performance significantly degrades ifthe videos are recorded under more challenging conditions,specifically when subjects’ motions and illumination varia-tions are involved. We propose a framework which utilizesface tracking and Normalized Least Mean Square adap-tive filtering methods to counter their influences. We testour framework on a large difficult and public databaseMAHNOB-HCI and demonstrate that our method substantially outperforms all previous methods. We also use our method for long term heart rate monitoring in a game evaluation scenario and achieve promising results.