2015 年的统计结果表明，心血管疾病是中国死亡人口的主要致死因素，分 别占农村和城市死亡人数的 44.60%和 42.51%，绝大多数心血管疾病都是因长期 患有高血压发展而成的[1]。便携连续地进行血压测量，对于预防和治疗高血压， 对心血管疾病的确诊治疗以及药物疗效的预后判断，都具有至关重要的作用[2]。 如今全自动数字血压计的使用已经非常普遍，极大地方便了人们进行血压监 控和测量。目前的数字血压计，大多数都是基于示波法的原理来进行血压的测量， 这种方法都需要通过袖带对人体上臂肱动脉进行压迫，从而测取血压。但此类血 压计存在以下不足:(1)只能间歇式地进行血压测量，无法获取连续的血压波动 曲线;(2)单次测量时间较长，期间个体运动会引入测量误差;(3)操作使用血 压计仍然需要一定的经验;(4)短时间内多次测量血压会导致手掌处供血不足而 发麻;(5)测量期间袖带限制了个体的活动，期间个体无法正常活动与工作。 本论文??出了一种基于人脸视频的血压测量方法，通过面部视频非接触式地 获取人体脉搏波波速(Pulse Wave Velocity, PWV)，进而依据 PWV 与人体收缩压 的准线性关系构建血压估计方程，最后通过 PWV 对人体收缩压进行估计。使用 该方法进行血压测量无需接触人体，可以实现连续的无干扰血压测量。本论文的 主要工作包括: 一、基于面部视频的脉搏波波速估计方法研究。对面部视频数据进行裁剪与 空间滤波获得迹曲线。为了从迹曲线中获得光电容积脉搏波，本论文引入盲源?? 取算法(Blind Source Extraction, BSE)，克服环境光变化和人脸抖动等噪声，实 现光电容积脉搏波信号(Photoplethysmography, PPG)??取。求取面部两个感兴 趣区域的两路 PPG 的相位差即可得到 PWV。然而，实验发现原有的 BSE 算法 收敛稳定性差、鲁棒性不强，针对该缺点本论文??出改进的 BSE 算法。实验结 果表明，??取结果与验证数据的相关系数为 0.8013，从视频数据中可以稳定可靠 地估计 PPG; 二、基于脉搏波波速的血压估计模型研究。完成 PPG 信号的??取之后求出 PWV，利用 PWV 可以对血压进行估计。本论文使用一元线性回归模型对血压进 行估计，并探讨了时间窗大小对回归结果的影响。结果表明本论文方法的血压估 计精度较高，平均绝对误差为5.7040 ± 0.9349mmHg，误差率为5.070% ± 0.78%。 通过 Bland-Altman 方法分析，本论文方法与传统血压测量设备具有很高的一致性。时间窗的选择过大或过小都会降低 PPG 信号的信噪比，结果表明时间窗的 大小为 4s 左右血压的回归结果最好，平均绝对误差为 6.72mmHg; 三、基于面部视频的血压测量系统。突破了非接触血压测量的技术难点之后， 本论文??出了非接触式血压测量的完整技术路线，并且基于 PYTHON 实现了非 接触式血压的测量系统，包括数据采集、数据处理、结果可视化等模块。在数据 采集模块，为了保证指端光电容积脉搏波数据与视频数据间的时间同步，利用蓝 牙控制器作为实验开始的触发器，设计了完善的数据采集流程以及完成了串口数 据传输模块代码开发;在数据处理模块，引入了目标跟踪与盲源??取等算法，?? 高 PPG ??取的可靠性;在可视化模块，基于 PYTHON 实现了人机交互界面以及 结果的可视化界面。 测试数据和真实数据计算结果表明:本文??出的改进盲源??取算法的信号?? 取结果优于目前普遍使用的盲源分离算法与基于图像的光电容积脉搏信号获取 技术，改进盲源??取算法的??取结果与验证信号具有更高的相关系数。通过回归 分析，结果表示基于面部视频的血压测量与真实血压结果具有很高的一致性。 本论文实现了基于面部视频的血压测量方法，实现了非接触式的心率的连续 测量，极大便利了人们的日常生活中与工作中的血压测量与监控。

It’s reported in 2015 that cardiovascular diseases are the main causes of death in China, accounting for 44.60% and 42.51% of the deaths in rural and urban areas, respectively. Long-term and effective monitoring and recording of blood pressure plays an important role in preventing hypertension and cardiovascular diseases, diagnosis and treatment of cardiovascular diseases. Nowadays, digital electronic sphygmomanometer is widely used, which greatly facilitates the spontaneous measurement of blood pressure. However, most of the digital electronic sphygmomanometers, both in hospitals and families, are designed on the principle of Korotkoff-Sound and oscilloscope to measure blood pressure, which need to compress the brachial artery of the upper arm through cuff to measure blood pressure. Current sphygmomanometers have the following shortcomings in application: (1) It only supports blood pressure measurement intermittently and cannot record blood pressure continuously; (2) Each measurement last for a long time, , and subject was require to stay quietly to avoid measurement errors from any small movement during blood pressure measurement; (3) Sphygmomanometers operate under the direction or guidance of the professional staff; (4) Blood pressure measurements repeat in a short time can lead to palm anesthesia due to compression of the brachial artery and insufficient blood supply; (5) The limitation of individual's activity during the measurement period. In this thesis, a novel blood pressure measurement method using face video is proposed. Here, Pulse Wave Velocity (PWV) is extracted from face video in a non- contact way, and then the blood pressure estimation equation is constructed to capture the quasi-linear relationship between PWV and systolic blood pressure. Thus, the systolic blood pressure is estimated from PWV. This method measures blood pressure without touching human body, which can realize continuous non-interference blood pressure measurement. First, we researched the extraction of pulse wave velocity estimation from face video. The face video data is cropped and spatially filtered to obtain the trace curve. In order to obtain the photoplethysmography (PPG) from the trace curve, we introduces blind source extraction (BSE) to overcome the noise of ambient light and face jitter, and realize the photoplethysmography extraction. PWV can be obtained by taking the phase difference of two PPG which obtain from two regions of interest on the face. To improve the stability and robustness of BSE algorithm, an improved BSE algorithm was introduced proposed. The experimental results show that the correlation coefficient between the extraction result and the verification data is 0.8013, and the PPG can be estimated stably and reliably from the video data. Second, the relation between blood pressure and pulse wave velocity was modeled with regression method. The PWV is obtained from the extracted PPG signal, and the blood pressure is then estimated from the PWV. Here, we used a linear regression model to estimate blood pressure from PWV, and explored the influence of time window length. The results show that the accuracy of blood pressure estimation by this method is high, the average absolute error is 5.7040±0.9349mmHg, and the error rate is 5.070%±0.78%. The Bland-Altman analysis results also prove the output blood pressure results is highly consistent with that of traditional blood pressure measurement equipments. Our results also demonstrated that if time window is too large or too small, the signal-to-noise ratio of PPG signal will be reduced, and the the optimal time window length is about 4 second, and the average absolute error is 6.72mmHg. Last, a blood pressure measurement system using face video was designed and implemented. this thesis proposed a complete technical solution for non-contact blood pressure measurement, and implemented a non-contact blood pressure measurement system code by PYTHON, including data acquisition module, data processing module, and visualization of results module. In the data acquisition module, in order to ensure the time synchronization between the fingertip photoplethysmography data and the video data, the Bluetooth controller was used as the trigger for the experiment, and the perfect data acquisition process and the serial port data transmission module code development were completed. In the signal processing module, algorithms such as target tracking and blind source extraction are introduced to improve the reliability of PPG extraction. In the visualization module, the human-computer interaction interface and the visual interface of the results are realized using PYTHON. Our Experimental results demonstrated that the improved blind source extraction algorithm outperforms traditional blind source separation algorithm and image-based photoelectric volume pulse signal acquisition technology. Simulation results showed that our methods improved the correlation coefficient between extracted wave and ground truth signal. The results also showed that the blood pressure of the regression model was highly consistence with that measure with professional equipment. This thesis introduce a blood pressure measurement method using facial video and its implementation, which can continuously measure the blood pressure in a non- contact manner. This studis greatly facilitates the blood pressure measurement and monitoring in people's daily life.