目的：研制基于无标记动作捕捉（Markerless Motion Capture，MMC）的三维动态定量测量装置，该装置可定量的输出面部各特征点在运动时的位移，作为患者面瘫程度评价的指标。

方法: 以人脸关键点的二维估计过程为基础，在三维人脸关键点重建中使用多个联动相机的姿态估计作为重建数据来源，通过相机的视野连续性及多目高精度重建技术，重建人脸关键点三维空间坐标，实现人脸关键点多目三维跟踪。利用3架联动相机（ZCAM E2，深圳视觉科技有限公司，深圳，中国）采集面神经功能障碍患者的面部表情数据，进行训练并实现面部特征点识别及三维重建。并通过与数控车床（CNC4380-800W，晶研仪器科技公司，东莞，中国）进行固定移动位移对比，来验证该装置精度。招募50例不同性别/年龄健康志愿者重复测量面部表情数据，用Bland-Altman分析图对两次测量数据进行一致性分析，验证三维动态定量测量装置的稳定性，并分析50例健康志愿者面部表情数据。

结果: 1. 完成了基于无标记动作捕捉的面部三维动态定量测量装置的搭建，可实时读取被测试者的面部运动数据，建立了以距离、角度为主要测量参数的面部运动报告表；2. 与数控车床（CNC4380-800W，晶研仪器科技公司，东莞，中国）进行固定移动位移对比，平均位移20 mm时，位移误差2 mm左右。3. 50例不同性别/年龄健康志愿者间隔大于24小时的重复测量，Bland-Altman分析图显示特征点重复测量位移值均有超过90%位于一致性界限（Limits of Agreement，LOA）内。

结论: 本研究研制的基于无标记动作捕捉的面部三维动态定量测量装置经过精度及稳定性验证，可以初步应用于患者面瘫程度的测量。

Objective: To develop a 3D dynamic quantitative measurement device based on Markerless Motion Capture (MMC), which can quantitatively output the displacements of facial feature points during movement as an index for evaluating the degree of facial paralysis in patients.

Methods: Based on the two-dimensional estimation process of facial key points, we use the pose estimation from multiple synchronized cameras as the data source for 3D facial key point reconstruction. Through the continuity of the camera's field of view and multi-view high-precision reconstruction technology, we reconstruct the 3D spatial coordinates of facial key points to achieve multi-view 3D tracking of facial key points. Three synchronized cameras (ZCAM E2, Shenzhen Vision Technology Co.Ltd. Shenzhen, China) are used to collect facial expression data from patients with facial nerve dysfunction for training and facial feature point recognition and 3D reconstruction. The accuracy of the device is verified by comparing the fixed movement displacement with a CNC lathe (CNC4380-800W, Jingyan Instrument Technology Co., Ltd., Dongguan, China). We recruited 50 healthy volunteers of different genders and ages to repeatedly measure facial expression data. The Bland-Altman analysis plot is used to analyze the consistency of the two sets of measurement data, verify the stability of the 3D dynamic quantitative measurement device, and analyze the facial expression data of 50 healthy volunteers.

Results: 1. We have successfully built a marker-less motion capture-based facial 3D dynamic quantitative measurement device that can real-time capture and read facial movement data from test subjects. We have also established a facial movement report sheet with distance and angle as the main measurement parameters. 2.  When comparing the fixed movement displacement with a CNC lathe (CNC4380-800W, Jingyan Instrument Technology Co., Ltd., Dongguan, China), the displacement error is less than 2 mm at an average displacement of 20 mm. 3. Through repeated measurements of 50 healthy volunteers of different genders and ages, with intervals greater than 24 hours, the Bland-Altman analysis chart shows that over 90% of the repeated measurement displacement values for feature points fall within the Limits of Agreement (LOA).

Conclusion: The facial three-dimensional dynamic quantitative measurement device based on markerless motion capture developed in this study has undergone verification of accuracy and stability, and can be preliminarily applied to measure the degree of facial paralysis in patients.