Abstract: Abstract: 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.

Key words: Keywords: point cloud, face alignment, mapping, feature extraction, spectacle frame fitting