Scholarlink

<p>Non-contact heart rate monitoring is increasingly important in remote care and infection control. Optical methods such as remote photoplethysmography (rPPG) and infrared thermography (IRT) can estimate heart rate but are easily affected by illumination, airflow, and temperature variations. In contrast, millimeter-wave (mmWave) radar has the advantages of being immune to light interference and low cost, and can estimate heart rate from chest movements. However, further suppression and compensation are needed to reduce its susceptibility to interference from factors such as respiratory drift, reflected clutter, and phase instability. This study proposes a time–frequency fusion framework that combines composite filtering and cepstrum analysis to enhance the stability of mmWave heart rate estimation. Noise is progressively suppressed using bandpass (BP), wavelet, median, and amplification (AMP) filtering, followed by drift compensation with <i>Peaks</i> + <i>Drift</i> and electrocardiogram (ECG) alignment. Heart rate is inferred by reconstructing periodic structures through the fast Fourier transform (FFT) and cepstrum, with peak tracking constrained within physiological ranges. Experiments comparing 12 filter sequences, parameter sensitivity, and noise-stress tests show that <i>AMP to BP</i> yields the best performance, achieving 4.3&#xa0;bpm MAE, 8.4&#xa0;bpm RMSE, and 0.8&#xa0;bpm bias at a 4-s window, 0.5-s hop, and <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\({q}_{\text{min}}\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>q</mi> <mtext>min</mtext> </msub> </math></EquationSource> </InlineEquation> = 0.5. Across six noise types, estimation error decreased as SNR increased, remaining lower than the raw signals. The overall estimation behavior is continuous and reproducible, indicating that this architecture can provide stable non-contact heart rate monitoring at a low cost and can serve as a reference for future clinical applications. The project of this work is made publicly available at <a href="https://github.com/seannnnnn1017/radar-heartbeat-detection">https://github.com/seannnnnn1017/radar-heartbeat-detection</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Millimeter-wave radar heart rate monitoring via time-frequency fusion and composite filtering with ECG alignment

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Chi Hung Wang,
  • Xiang Shun Yang,
  • Yu Hsiang Hsiang

摘要

Non-contact heart rate monitoring is increasingly important in remote care and infection control. Optical methods such as remote photoplethysmography (rPPG) and infrared thermography (IRT) can estimate heart rate but are easily affected by illumination, airflow, and temperature variations. In contrast, millimeter-wave (mmWave) radar has the advantages of being immune to light interference and low cost, and can estimate heart rate from chest movements. However, further suppression and compensation are needed to reduce its susceptibility to interference from factors such as respiratory drift, reflected clutter, and phase instability. This study proposes a time–frequency fusion framework that combines composite filtering and cepstrum analysis to enhance the stability of mmWave heart rate estimation. Noise is progressively suppressed using bandpass (BP), wavelet, median, and amplification (AMP) filtering, followed by drift compensation with Peaks + Drift and electrocardiogram (ECG) alignment. Heart rate is inferred by reconstructing periodic structures through the fast Fourier transform (FFT) and cepstrum, with peak tracking constrained within physiological ranges. Experiments comparing 12 filter sequences, parameter sensitivity, and noise-stress tests show that AMP to BP yields the best performance, achieving 4.3 bpm MAE, 8.4 bpm RMSE, and 0.8 bpm bias at a 4-s window, 0.5-s hop, and \({q}_{\text{min}}\) q min = 0.5. Across six noise types, estimation error decreased as SNR increased, remaining lower than the raw signals. The overall estimation behavior is continuous and reproducible, indicating that this architecture can provide stable non-contact heart rate monitoring at a low cost and can serve as a reference for future clinical applications. The project of this work is made publicly available at https://github.com/seannnnnn1017/radar-heartbeat-detection.