Rendering

A simple approach to image-based rendering is presented in a paper by Jan Kautz and Hans-Peter Seidel called Hardware Accelerated Displacement Mapping for Image Based Rendering. This paper was published in Graphics Interface in 2001. Much of the paper is dedicated to 3D reconstruction of objects based on depth map images. Depth map images are not applicable to the usual case of rendering straight from cameras, but if you look at section 5 of the paper it explains how the depth map technique can be applied to the simpler case of Image-based visual hulls. This is the simplest approach to rendering a visual hull from image-based inputs that I have come across. The approach is to load each of the textures (from real cameras) into texture units on the graphics card. You can then use register combiners to combine the projected textures on the geometry. The geometry in this approach are polygonal slices orthogonal to the view plane of the virtual camera.

I use the rendering technique proposed by Ming Li, Marcus Magnor and Hans-Peter Seidel in the paper titled: Hardware-Accelerated Visual Hull Reconstruction and Rendering. This paper was published in Graphics Interface in 2003. The difference between this approach and the approach above is that Li calculates more accurate geometry for which to project the textures onto. This approach should lead to a nicer reconstruction, but the improvement in practice with only a few cameras is negligible. The amount of extra effort involved in this approach is hard to justify and so I recommend the Kautz approach above.

Both of these papers were published before the popularization of shading languages such as GLSL or Cg. If you are interested in implementing a 3D reconstruction approach I suggest using Kautz's image-based technique but instead of using register combiners to calculate texel values use a shader language such as GLSL or CG.