| Title |
Predicting Heart Rate Using 3D-CNN and Wavelet Transform from rPPG |
| Authors |
김태완(Tae-Wan Kim) ; 곽근창(Keun-Chang Kawk) |
| DOI |
https://doi.org/10.5370/KIEEP.2023.72.4.315 |
| Keywords |
3D Convolutional neural network; CWT algorithm; photolethysmography; heart rate; deep learning |
| Abstract |
In recent years, research has been conducted into technologies that acquire biological signals in various situations such as healthcare and smart cars, and analyze the acquired signals to perform user recognition, emotion classification, and health status monitoring.
Among them, heartbeat is a direct element that represents a person's physical and mental state. Conventional photoplethysmography (PPG) and electrocardiogram (ECG) methods require separate hardware for acquisition, making it difficult to acquire heartbeats in everyday situations. In this paper, we pre-processing videos to emphasize parts related to blood flow and reduce computing resource and times. and we design a model that utilizes 3D-CNN (3Dimension-Convolutional Neural Networks) to predict blood flow signals in a non-contact method through one second facial videos, while reducing the number of input channels. The blood flow signal predicted from the learned model extracted frequency regions related to heartbeat in various frequency regions through CWT (Continuous Wavelet Transform) analysis, and predicts heart rate through the frequency with the highest energy signal among them.
Finally, we compared the predicted heart rate from one second facial videos with heart rates obtained though sensors, verified the proposed pre-processing methods and model’s performance |