Reconstructing Three-dimensional Models from Multi-View Images Using BSC Method and Its Evaluation
Details
Methods which reconstruct 3D models from multi-view images come to the front due to recent advancement in image processing. There is a matter to be incapable of concavo-convex shape with spike in convention methods reconstructing 3D models from multi-view images. This paper proposes a method reconstructing 3D models from multi-view images of the object with concavo-convex shape and investigates a method to evaluate its models The proposed method calculates a field including proper surface of the object using Block Space Carving method and deforms the model in it. In deformation, the proposed method calculates a base surface of the object and deforms the model as a center of it. The proposed method reconstructs 3D models more accurate than conventional methods, by this processing. After deformation, the proposed method synthesizes 2D images at the camera positions the captured photographs and estimates accuracy of the 3D models. Replicating the deformation and the estimation accuracy of the 3D models, the proposed method reconstructs 3D models considering concavo-convex shape. At the end of this paper, the proposed method runs simulation experiments and experiments with photographed images. In the simulation experiments, the investigated method compares voxel data reconstructed from multi-view images with voxel data made from CG models data of the objects. In the experiments with photographed images, this paper evaluate the 3D models using PSNR. These results show that the proposed method is effective.
