基于面部视频的低运算量心跳检测  

Low-Computation heartbeat detection based on facial video

作　　者：丁国栋 鞠明 于坤灿 向新[3] DING Guodong;JU Ming;YU Kuncan;XIANG Xin(The School of Computer Science,Xi′an Aeronautical University,Xi′an,Shaanxi 710077,China;Unit 93131 of People′s Liberation Army,Beijing 100000,China;Aviation Engineering School,Air Force Engineering University,Shaanxi 710038,China)

机构地区：[1]西安航空学院计算机学院,陕西西安710077 [2]93131部队,北京100000 [3]空军工程大学航空工程学院,陕西西安710038

出　　处：《燕山大学学报》2024年第3期236-243,共8页Journal of Yanshan University

基　　金：陕西省自然科学基金资助项目(2021JM-220)。

摘　　要：基于面部视频的心跳检测能够提升面部识别技术的可信度,在军事领域中具有对抗面部伪装、判断真假人目标、鉴定伪造视频等应用潜力。然而,现有基于视频的心跳检测算法存在运算量较大的问题,限制了其在低运算能力条件下的实时应用。本文分析了算法原理和关键步骤的时延,并提出了两种低运算量的基于插值运算的算法,分别是RGB-FFT算法和RGB-ICA-FFT算法。经过测试,在帧速率为5 fps的情况下,插值RGBFFT算法的运算量最小;插值RGB-ICA-FFT算法在精度上与帧速率为15 fps的RGB-ICA-FFT算法相当,但运算量低61%。Heartbeat detection based on facial videos can enhance the credibility of facial recognition technology,enabling applications in military domains such as countering facial disguises,distinguishing between real and fake targets,and identifying forged videos.However,existing video-based heartbeat detection algorithms suffer from high computational complexity,limiting their real-time applicability under low computational capabilities.In this paper,the algorithm principles and the time delay of key steps are analyzed,and two low-complexity algorithms based on interpolation are proposed:the RGB-FFT algorithm and the RGB-ICA-FFT algorithm.Through testing,it is found that the interpolated RGB-FFT algorithm has the lowest computational complexity at a frame rate of 5 fps.The interpolated RGB-ICA-FFT algorithm achieves comparable accuracy to the RGB-ICA-FFT algorithm at a frame rate of 15 fps,but with a 61%reduction in computational complexity.

关 键 词：心跳检测 人脸识别 独立成分分析 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]