智能配镜三维特征参数提取方法研究

Feature Extraction of Facial Point Cloud Data for Intelligent Spectacle Frame Fitting

针对智能配镜中三维面部特征点提取算法复杂度较高的问题,提出一种将三维点云转换为映射图像定位特征点的方法。采用Voronoi方法计算面部三角网格各顶点处的高斯曲率、平均曲率。选取鼻尖、眼角等曲率特征明显的区域估计面部点云姿态。根据曲率旋转不变性,使用初选的点云方向向量简化旋转矩阵的计算,使面部点云正面朝向视点。将点云映射转换为图像,三维网格模型中三角面片一对一映射到图像中的三角形。搭建卷积神经网络,使用Texas 3DFRD数据集进行模型训练。进行人脸对齐,预测所得各面部特征点分别限制在图像某三角形中。根据图像中三角形映射查找三维网格模型中对应三角面片,通过三角面片顶点坐标计算配镜所需的面部特征点位置坐标,实现配镜特征参数的提取。

In order to extract the 3D facial point cloud in the intelligent spectacle frame fitting,a method of converting triangulated 3D point cloud into a mapped image is proposed to locate feature points.Firstly,the Gaussian curvature and mean curvature of each vertex of the face triangular mesh are calculated by Voronoi method.Secondly,nasal tip and canthus regions with obvious curvature features are used to estimate the facial point cloud orientation.According to the rotation invariance of surface curvature,the calculation of rotation matrix is simplified by using the point cloud orientation vector,so that the face point cloud faces the viewpoint.Then,the point cloud is transformed into an image by mapping,and the triangular faces in the 3D mesh model are mapped to the biunique triangles.The convolutional neural network is built and the Texas 3D Face Recognition Database is used for model training.Finally,face alignment is carried out and the predicted facial feature points are limited to a certain triangle of the image.According to the triangle mapping in the image,the corresponding triangular faces in the 3D mesh model can be found,and the coordinates of the facial feature points required by the spectacle frame fitting are calculated through the vertex coordinates of the triangular faces.Through the above steps,the feature extraction of spectacle frame fitting parameter is implemented.