Research On The Application Of Neural Network To Surface Reconstruction In Reverse Engineering

Reconstructing digital models of physical existing objects by means ofcoordinate measuring is usually called reverse engineering (RE) which iswidely used in a variety of fields, such as mechanism, iatrology, movie etc.Surface reconstruction is a key technology in RE, it is a new project to applyartificial neural network to surface reconstruction in RE. Unorganized clouddata simplification is realized, applying Self-Organizing Feature Map (SOM)neural network to surface reconstruction and Radial Basis Function (RBF)neural network to hole repairing of the triangular mesh model are researchedin this paper. The main researches and achievements as follows: (1) The simplification technology of unorganized cloud data isresearched. Based on the cluster result of the unorganized cloud data, twosimplification methods, based on the nearest distance and based on MovingLeast-Squares (MLS), are put forward and compared in running time andsimplification effect. (2) The approach of applying SOM neural network to surfacereconstruction is studied. A novel initialization method of SOM and a newalgorithm for segmentation study to scattered data points are proposed, whichare aiming at facilitating the process time of computer and improving thenetwork study effect remarkably. (3) The method of hole repairing in triangular mesh surfaces based onRBF neural network is established. A novel algorithm for hole repairing ispresented and implemented, the RBF network is trained by using the trianglevertices around the hole, the output of network accurately describes thesurface function expression of hole field, the network precision is improved.