Learning-based Remote Photoplethysmography for Physiological Signal Feedback Control in Fitness Training

Remote photoplethysmography (rPPG) has attracted much attention in recent years. This research proposes to apply rPPG to the fitness training scenario, enabling non-contact measurement of the subject's heart rate during training. Currently, most existing approaches suffer from a major weakness, i.e. the subject's body needs to remain stably while conducting measurement, which significantly hinders practical applications of the approach. The main purpose of this paper is to build a training system based on rPPG and fitness machines to provide users with better ergonomic exercise experiences. We have built a spinning bike system that combines a camera and an adaptive controller based on heart rate feedback for tracking the desired exercise intensity. Fuzzy control is introduced in the feedback control loop by considering heart rate and heart rate variability simultaneously for better representation of the physical status. Some preliminary results are briefly presented. This research demonstrates promising performance improvement by combining rPPG heart rate estimation and fitness machine control.