Visual heart rate estimation and negative feedback control for fitness exercise

Abstract

The heart rate of subjects is extracted from video data during exercises and is treated as a negative feedback parameter for the control of fitness devices. The probability distribution in RGB space of the parameters of the skin reflection model is analyzed based on the experimental data taken under a wide range of environmental settings. A novel pulse extraction algorithm is proposed based on this distribution, which sets aside distribution centers as the model parameters for preliminary pulse separation and adopts cascade least mean square adaptive filters to compensate for modeling errors. A Gaussian distribution-based heart rate estimation algorithm is employed for heart rate stabilization. In addition, the concept of heart rate negative feedback control is introduced, e.g., the treadmill speed and incline can be adjusted interactively based on the subject's heart rate. The heart rate measurement accuracy and the control scheme have been evaluated by a self-collected benchmark dataset and a prototype system. The heart rate measurement accuracy is increased over the state-of-the-art algorithms by more than 10% on average. The heart rate negative feedback control improves human–machine interaction performance compared with a traditional control scheme that uses the predefined program regardless of the subjects’ physical state. The proposed technique pioneers new possibilities for the usage of fitness devices.