基于运动特征预估的呼吸率检测快速算法

A fast algorithm for respiratory rate detection based on motion feature estimation

呼吸率在呼吸疾病的预防与诊断上有重要意义。现有基于视频的呼吸率检测算法复杂度较高,会造成嵌入式实现延时较大的问题。针对该问题,文章提出一种基于运动特征预估的呼吸率检测快速算法,其核心思想是分析呼吸率信号的主要频率、幅度、方向,以实现感兴趣信号的预估和分析。首先,通过人脸检测确定人脸区域,结合人体比例选出胸部区域;其次,设计一种基于运动特征预估的呼吸信号提取方法,利用最大似然估计呼吸基频,利用基频计算幅度并选取呼吸区域,利用方向梯度直方图获取呼吸运动主要方向;接着,在获取呼吸区域和运动方向后,采用基于相位的欧拉视频处理算法提取呼吸信号;最后,利用傅里叶变换的峰值检测获取呼吸率。实验表明,与现有算法相比,该方法能够有效减少60%的运行时间,具有更好的实时性,且在实验室环境下呼吸率平均误差为0.55次/min,与真实呼吸率具有良好的一致性。

Respiratory rate is of great significance in the prevention and diagnosis of respiratory diseases.The existing video-based respiratory rate detection algorithms have high complexity,which may cause large delay in embedded implementation.In response to this problem,a fast algorithm for respiratory rate detection based on motion feature estimation is proposed.The core idea is to analyze the main frequency,amplitude and direction of respiratory rate signal to achieve the estimation and analysis of the signal of interest.Firstly,the face area is determined by face detection,and the chest area is selected based on the proportion of human body.Secondly,a respiratory signal extraction method based on motion feature estimation is designed.In this method,the maximum likelihood is used to estimate the respiratory fundamental frequency,the respiratory region is selected by the amplitude calculated by the fundamental frequency,and the main direction of respiratory motion is obtained by the histograms of oriented gradients(HOG).After acquiring the respiratory region and motion direction,the phase-based Eulerian video processing algorithm is used to extract the respiratory signal.Finally,the peak value detection of Fourier transform is used to obtain the respiratory rate.Experiment shows that compared with the existing algorithms,the proposed method can effectively reduce the running time by 60%and has better real-time performance.At the same time,the average error of the respiratory rate in the laboratory environment is 0.55 times/min,which is in good agreement with the real respiratory rate.