基于光流的非接触式呼吸频率检测算法研究

Research on a Non-Contact Respiratory Rate Detection Algorithm Based on Optical Flow

摘要:

针对传统接触式呼吸频率检测存在舒适性差、应用场景受限等问题，对基于普通RGB视频的非接触式呼吸频率检测算法进行了研究；采用图像梯度光流构建人体呼吸区域运动光流场，通过累加张量与方向投影提取呼吸信号，并结合Savitzky-Golay滤波和短时傅里叶变换实现呼吸频率估计；为验证算法性能，开展了人体多姿态实验以及衣物厚度、衣物纹理和视频时长等多场景稳定性实验，并与两种典型非接触式呼吸频率检测算法进行了对比分析；实验结果表明，所提出算法在正卧、侧卧、正坐和侧坐等姿态下均实现了较高检测精度，平均绝对误差小于1 bpm，检测准确率超过95%，在复杂场景及长时间连续监测条件下仍保持良好的稳定性和鲁棒性；研究结果表明，该算法能够实现准确、稳定的非接触式呼吸频率检测，能够满足日常环境下的呼吸频率检测需求。

Abstract:

To address the problems of poor comfort and limited application scenarios in traditional contact-based respiratory rate detection methods, a non-contact respiratory rate detection algorithm based on ordinary RGB video is studied. An optical flow field of the human respiratory region is constructed using image gradient optical flow, and the respiratory signal is extracted through optical flow accumulation tensor and directional projection. The Savitzky–Golay filter and short-time Fourier transform (STFT) are combined to estimate the respiratory rate. To verify the performance of the proposed algorithm, multi-posture experiments and multi-scenario stability experiments involving clothing thickness, clothing texture, and video duration are conducted, and two typical non-contact respiratory rate detection algorithms are se-lected for comparative analysis. The experimental results show that the proposed algorithm achieves high detection accuracy under different postures, including supine, lateral lying, upright sitting, and lateral sitting positions. The mean absolute error is less than 1 bpm, and the detection accuracy exceeds 95%. The algorithm maintains good stability and robustness under complex scenarios and long-term continuous monitoring conditions. The results indicate that the proposed algorithm can achieve accurate and stable non-contact respiratory rate detection and meet the requirements of respiratory rate detection in daily environments.