Vision-Based Lightweight Facial Respiration and Heart Rate Measurement Technology

Heart rate and respiratory rate are important information reflecting human vital signs. Many studies have used face detection and feature point tracking to find ROI (region of interesting) for non-contact analysis and measurement. However, these methods all convert the features into a single signal for processing, and when there is noise interference in the ROI area, it is easy to cause measurement errors. This paper adopts multiple ROIs to select multiple features. Significant signals are selected by correlation and integrated by principal component analysis to reduce the impact on measurements when individual signals are disturbed. Considering that the real-time measurement is affected by the computational performance, this paper uses optical flow tracking to replace the face detection of each frame which makes the tracking features more stable. This paper measures the RPPG (remote photoplethysmography) signal to estimate the heart rate, and then measures the respiration according to the facial micro-vibration generated by the breathing movement. Experimental results show that for heart rate measurement, the proposed method achieves a MAE of 6.53 and an RMSE of 8.684 in the Ostankovich’s dataset. Furthermore, the proposed method is combined with the ensemble empirical mode decomposition, and the MAE and RMSE are further reduced to 4.124 and 6.897. The respiration rate detection results of the proposed method have a MAE of 2.017 and an RMSE of 2.676. Furthermore, the proposed method can actually run on a Raspberry Pi 4 platform with less computing power, proving the real-time processing capability of the system.