Abstract

Info

Title:: Non-contact physiological parameter detection method based on improved three-dimensionalconvolution network

Author(s):: XU Zhanyu; CHEN Zhaoxue; School of Health Science and Engineering, University of Shanghai for Science and Technology, Shanghai 200093, China

Keywords:: non-contact type; heart rate detection; hybrid attention mechanism; signal processing

Abstract:: Remote photoplethysmography is a method of measuring physiological parameters such as heart rate from facialvideo. For overcoming the difficulties in achieving both high accuracy and lightweight by the existing heart rate measurementmethods, an improved three-dimensional convolution network model is proposed to realize non-contact physiologicalparameter detection in facial video. In the pre-processing, YuNet model takes place of the traditional face detector, so that theface region can be recognized quickly and accurately. In addition, attention mechanisms and residual modules are embed intothree-dimensional convolution network to extract key channel and spatial features, with long short-term memory networksused as period memory modules to capture long-term dependencies in the data. The experimental results show that theproposed Res-CHATM model achieves excellent results of MAE=2.19 BPM, RMSE=7.02 BPM, C=0.95, and MAE=1.65BPM, RMSE=3.44 BPM, C=0.98 in the cross experiments on public datasets UBFC-rPPG and PURE for heart rateestimation. The consistency between the predicted value and the real value and the effectiveness of the fusion module arefurther verified, demonstrating the potential of efficient lightweight model in remote photoplethysmography.