A Method for Surface Reconstruction Based on Support Vector Machine

Surface reconstruction is one of the main parts of reverse engineering and environment modeling. In this paper a method for reconstruct surface based on Support Vector Machine (SVM) is proposed. In order to overcome the inefficiency of SVM, a feature-preserved nonuniform simplification method is employed to simplify cloud points set. The points set is reduced while the feature is preserved after simplification. Then a reconstruction method based on segmented data is proposed to accelerate SVM regression process for cloud data. Firstly, the original sampling data set is partitioned to generate several training data subsets and testing data subsets. A segmentation technique is adopted to keep the continuity on the borders. Secondly regression calculation is executed on every training subset to generate a SVM model, from which a segmented mesh is obtained according to the testing data subset. Finally, all the mesh surfaces are stitched into one whole surface. Both theoretical analysis and experimental result show that the segmentation technique presented in this paper is efficient to improve the performance of the SVM regression, while keeping the continuity of the subset borders. Index Terms — Surface reconstruction, surface variance, support vector machine, segmentation