Automatic determination of the normal posture of 3D objects and the superposition of 3D objects using deep learning

摘要

The present invention relates to a method for automatically determining a normal posture of a 3D object represented by a 3D data set, the method comprising: a first 3D neural network is a normal posture associated with a normal coordinate system defined for some positions of a 3D tooth structure. Providing at least one block of voxels representing voxels of a 3D object associated with the first coordinate system, trained to generate information, to an input of the first 3D deep neural network; The normal posture information includes prediction of the data point position in the normal coordinate system for each data point of one or more blocks, the data point position is defined by normal coordinates, and the normal posture information is obtained from the output of the first 3D deep neural network. Receiving by the processor; Orientation and scaling of the normal coordinate system axis using normal coordinates, and a transformation parameter for determining the position of the normal coordinate system with respect to the first 3D coordinate system origin and axis, and converting the coordinates of the first coordinate system to normal coordinates using the orientation and position Determining; And, determining a normal representation of the 3D tooth structure, the determination comprising applying a transform parameter to the voxel coordinates of the voxel representation or the coordinates of the 3D data set used to determine the voxel representation.