中文 | English
Hefei University of Technology Logo J. Hefei Univ. Tech. (Nat. Sci.)

Navigation

Non-contact heart rate detection based on weight optimized convolutional neural network

Download Full Text

基于权重优化卷积神经网络的非接触心率检测

Journal Information

J. Hefei Univ. Tech. (Nat. Sci.), Apr.2024, Vol.47 No.4: 479-487

DOI: 10.3969/j.issn.1003-5060.2024.04.008

Authors

WANG Panru $ ^{1,2} $, YANG Xuezhi $ ^{2,3} $, LIU Xuenan $ ^{2,3} $, LI Longwei $ ^{4} $, WANG Dingliang $ ^{4} $

(1, School of Computer Science and Information Engineering, Hefei University of Technology, Hefei 230601, China; 2, Anhui Province Key Laboratory of Industry Safety and Emergency Technology, Hefei 230601, China; 3, School of Software, Hefei University of Technology, Hefei 230601, China; 4, Department of Cardiovascular, The First Affiliated Hospital of University of Science and Technology of China (Anhui Provincial Hospital), Hefei 230002, China)

Abstract and Keywords

Abstract: Heart rate is an important indicator of the health and exercise status of the human body. Due to the limitations of traditional heart rate detection, many non-contact detection methods have been proposed in recent years, which have good results under cooperative conditions. But the accuracy is significantly reduced when there is motion disturbance. For this problem, this paper combines computer vision and deep learning frontier theory to propose a non-contact heart rate detection method based on weight optimized convolutional neural network. By optimizing the network structure of the convolutional neural network, the noise immunity performance of the network is improved, and a more accurate heart rate value is obtained. Firstly, the stable facial video is input. Secondly, the pixel average of the input facial video frame is taken by frame and row by row, and the time domain expansion is made to obtain each row of sub-pulse waves. The pulse matrix composed of the sub-pulse waves is processed by using principal component analysis (PCA) and band-pass filter. Finally, the resulting feature matrix is input into the weight optimized convolutional neural network to learn to predict the heart rate. To verify the performance advantages of this algorithm, 2 200 face video samples from the self-collected dataset were used for experimental analysis. The experimental results showed that the

Keywords: non-contact; principal component analysis(PCA); weight optimization; convolutional neural network; heart rate

作者信息

王盼孺 $ ^{1,2} $，杨学志 $ ^{2,3} $，刘雪南 $ ^{2,3} $，李龙伟 $ ^{4} $，王定良 $ ^{4} $

（1. 合肥工业大学 计算机与信息学院，安徽 合肥 230601；2. 工业安全与应急技术 安徽省重点实验室，安徽 合肥 230601；3. 合肥工业大学 软件学院，安徽 合肥 230601；4. 中国科学技术大学第一附属医院（安徽省省立医院）心血管内科，安徽 合肥 230002）

摘要和关键词

摘要: 心率是反映人体健康状况和运动状态的重要指标。针对传统心率检测的局限性,近年来提出了许多非接触式的检测方法,其在配合条件下的效果较好,但是在有运动干扰时准确率明显下降。针对该问题,文章结合计算机视觉与深度学习前沿理论,提出一种基于权重优化卷积神经网络的非接触心率检测方法,通过优化卷积神经网络的结构,提高网络的抗噪性能,获得更准确的心率值。首先输入相对稳定的面部视频,然后将输入的面部视频逐帧按行取像素平均值并时域扩展得到各行子脉搏波,再使用主成分分析(principal component analysis,PCA)法与带通滤波器对各行子脉搏波组成的脉搏矩阵进行处理,最后将所得特征矩阵输入权重优化卷积神经网络学习,预测心率值。为了验证该方法的性能优势,使用自采数据集中的2 200份人脸视频样本进行实验分析,实验结果表明,文中所提方法与现有的非接触心率检测方法相比,具有更高的准确率、更强的鲁棒性。

关键词: 非接触；主成分分析(PCA)；权重优化；卷积神经网络；心率

Funding

安徽省科技重大专项资助项目(201903C080200100);安徽高校协同创新资助项目(GXXT-2019-003)和智能互联系统安徽省实验室(合肥工业大学)资助项目(PA2021AKSK0111)

Previous Next
Back to Contents

Account

Submission Login Reviewer Login Editor Login