Multi-resolution space carving using level set methods

Abstract

We present a multi-resolution space carving algorithm that reconstructs a 3D model of visual scene photographed by a calibrated digital camera placed at multiple viewpoints. Our approach employs a level set framework for reconstructing the scene. Unlike most standard space carving approaches, our level set approach produces a smooth reconstruction composed of manifold surfaces. Our method outputs a polygonal model, instead of a collection of voxels. We texturemap the reconstructed geometry using the photographs, and then render the model to produce photo-realistic new views of the scene.

Accurate rough terrain estimation with space-carving kernels

“…Space carving algorithms, originally suggested by Kutulakos and Seitz [10], use visibility information to produce a voxel model from calibrated images of a scene (a survey of space-carving approach can be found in [21]), but this strategy has not been adopted by the ladar community. Another approach that exploited the visibility constraints was the locus method of Kweon et al [11].…”

“…A fixed lengthscale with a compactly supported kernel function does not give global support. Many solutions have been proposed for this problem in the graphics community, where radial basis functions are used to estimate 3D surfaces and compactly supported kernels are chosen for efficiency [19,25,21]. Most commonly, backfitting, a general algorithm that can fit any additive model, is used to decrease the lengthscale incrementally [4].…”

Multiview stereo object reconstruction with a one-line search method

“…In voxel-based methods, space carving, [18][19][20][21] levelsets, [22][23][24] and surface mesh grid optimization [25][26][27] are typically exploited. These methods start with an initial surface guess larger than the object to reconstruct, which could be the visual hull or just a bounding box.…”

Mesh based 3D shape deformation for image based rendering from uncalibrated multiple views

“…The key issues of those approaches are to improve image quality with variant information derived from input images, because the differences between the reconstructed Visual Hull and the real object may cause blurring in the synthesized virtual images. The Space Curving Method [1] removes unnecessary voxels of the voxel-represented model using texture information by reducing difference between constructed model. The techniques for optimizing 3D model by deforming the vertex of surface triangle mesh based on the correlation of the texture have also been proposed [2] [3].…”