Image-based 3D modeling is the problem of recovering
scenes’ 3D geometry and appearance from images. Nowadays,
with the prevalence of digital cameras, image-based
3D modeling has the clear advantage over other 3D modeling
techniques in terms of equipment availability, affordability
and amount of user input, besides advantages on operation
conditions, scalability, etc.

The scene’s geometry and appearance is modeled with
an evenly divided discrete volume, each unit being a voxel.
Each voxel has two properties: opacity and color. Opacity
is a binary variable, 0 for empty and 1 for solid. The color of
the solid voxel represents the appearance of the object.

This algorithm uses a novel approach for multiview
stereo based on a Markov Random Field (MRF) model with ray-cliques modeling
the image formation process. With a multi-view image generation
probabilistic mixture model, and optimized belief
propagation, the MRF model can be inferred efficiently.

We are currently using 2 3D alogrithms.


Automatic generation of 3d models from photographs using Shubao Liu's iray software developed at the LEMS lab at Brown University.


Automatic generation of 3d models from photographs using Yasutaka Furukawa's Patch-based Multi-view Stereo software, PMVS-2, version 2, developed at the University of Washington and distributed under the GPL.