School of Physics, Northeast Normal University, Changchun 130022, China; Faculty of Physics, University of Science, Pyongyang, Democratic People's Republic of Korea.
Academy of Ultramodern Science, Kim Il Sung University, Pyongyang, Democratic People's Republic of Korea.
Comput Methods Programs Biomed. 2021 Mar;200:105824. doi: 10.1016/j.cmpb.2020.105824. Epub 2020 Nov 2.
The imaging photoplethysmography method is a non-contact and non-invasive measurement method that usually uses surrounding illumination as an illuminant, which can be easily influenced by the surrounding illumination variations. Thus, it has a practical value to develop an efficient method of heart rate measurement that can remove the interference of illumination variations robustly.
We propose a novel framework of heart rate measurement that is robust to illumination variations by combining singular spectrum analysis and sub-band method. At first, we extract the blood volume pulse signal by applying the modified sub-band method to the raw facial RGB trace signals. Then the spectra for the interference of illumination variations are extracted from the raw signal obtained from facial regions of interest by using singular spectrum analysis. Finally, we estimate the more robust heart rate through comparison analysis between the spectra of the extracted blood volume pulse signal and the illumination variations.
We compared our method with several state-of-the-art methods through the analysis using the self-collected data and the UBFC-RPPG database. Bland-Altman plots and Pearson correlation coefficients pointed out that the proposed method could measure the heart rate more accurately than the state-of-the-art methods. For the self-collected data and the UBFC-RPPG database, Bland-Altman plots showed that proposed method caused better agreement with 95% limits from -4 bpm to 10 bpm and from -2 bpm to 4 bpm respectively than the other state-of-the-art methods. It also revealed that the highly linear relation was held between the estimated heart rate and ground truth with the correlation coefficients of 0.89 and 0.99, respectively.
By extracting illumination variation directly from the facial region of interest rather than from the background region of interest, the proposed method demonstrates that it can overcome the drawbacks of the previous methods due to the illumination variation difference between the background and facial regions of interest. It can be found that the proposed method has a relatively good robustness regardless of whether illumination variation exists or not.
影像光体积描记法是一种非接触、非侵入式的测量方法，通常使用周围光照作为光源，但这种方法很容易受到周围光照变化的影响。因此，开发一种能够有效去除光照变化干扰的心率测量方法具有实际意义。
我们提出了一种新的心率测量框架，该框架通过奇异谱分析和子带方法相结合，可以稳健地去除光照变化的干扰。首先，我们通过应用改进的子带方法从原始面部 RGB 迹线信号中提取血液体积脉搏信号。然后，通过奇异谱分析从感兴趣面部区域的原始信号中提取光照变化的光谱。最后，通过比较分析提取的血液体积脉搏信号和光照变化的光谱，估计更稳健的心率。
我们通过使用自我收集的数据和 UBFC-RPPG 数据库对几种最先进的方法进行了分析，比较了我们的方法。Bland-Altman 图和 Pearson 相关系数表明，与最先进的方法相比，该方法可以更准确地测量心率。对于自我收集的数据和 UBFC-RPPG 数据库，Bland-Altman 图显示，与其他最先进的方法相比，该方法引起的偏差更好，在-4 bpm 到 10 bpm 和-2 bpm 到 4 bpm 的范围内分别为 95%的限制。它还表明，估计心率与真实心率之间存在高度线性关系，相关系数分别为 0.89 和 0.99。
通过直接从感兴趣的面部区域提取光照变化，而不是从感兴趣的背景区域提取光照变化，该方法表明它可以克服由于感兴趣的背景和面部区域之间的光照变化差异而导致的先前方法的缺点。可以发现，无论是否存在光照变化，该方法都具有相对较好的鲁棒性。
