4种基于rPPG的非接触式心率估计方法的性能评估

Performance evaluation of four non-contact heart rate estimation methods based on rPPG

目的:评估2SR(spatial subspace rotation)、POS(plane-orthogonal-to-skin)、Project_ICA和联合盲源分离与集合经验模态分解(joint blind source separation and ensemble empirical mode decomposition,JBSS_EEMD)4种具有代表性的基于远程光电容积脉搏波描记术(remote photoplethysmography,rPPG)的心率估计方法的性能。方法:采集黄种人和黑种人受试者的面部视频用于估计心率,同时用透射式手指脉搏血氧仪采集受试者的参考心率,在静止状态下和运动恢复状态下分别用4种方法估计受试者的心率,并与参考心率进行对比。结果:实验结果表明,在黄种人组处于静止和运动恢复状态以及黑种人组处于静止状态3种场景下,POS和Project_ICA方法的心率估计效果优于2SR和JBSS_EEMD方法。但在黑种人组处于运动恢复状态的场景下,4种方法的心率估计效果都较差。结论:在黄种人组处于静止和运动恢复状态以及黑种人组处于静止状态的场景下,引入皮肤反射模型,可以有效提高基于rPPG的非接触式心率估计方法的准确性。

Objective To evaluate the performance of four newly representative heart rate estimation methods based on rPPG:spatial subspace rotation(2SR)method,plane-orthogonal-to-skin(POS)method,Project_ICA method and joint blind source separation and ensemble empirical mode decomposition(JBSS_EEMD)method.Methods Facial videos of yellow and black subjects were collected for heart rate estimation,and a reference heart rate was collected using a transmission finger pulse oximeter,estimated using four methods at rest and at exercise recovery,and compared with the reference heart rate.Results The experimental results showed that the heart rate estimation by POS and Project_ICA methods were better than those by 2SR and JBSS_EEMD methods in the yellow subjects at rest and at exercise recovery scenarios,and also fits for the black subjects at rest scenario.However,the heart rate estimation by all the four methods was not good for the black subjects at exercise recovery scenario.Conclusion By comparing the methods of recent years,it is found that the introduction of skin reflection model can effectively improve the accuracy of the rPPG-based non-contact heart rate estimation for the yellow subjects at rest and at exercise recovery scenarios,and black subjects at rest scenario.