A Volumetric Reconstruction Method from Multiple Calibrated Views using Global Graph Cut Optimization

In this paper, a new formulation and solution to volumetric reconstruction from multiple calibrated images is presented. This problem has been previously formulated either as a continuous geometric optimization process driven by local numerical methods, or as a discrete labelling problem solved by global techniques for computing only stereo disparities. Our new formulation builds a bridge between these two approaches and takes advantage of both: a continuous geometric functional is minimized up to a discretization by a global graph cut algorithm. The relation between the continuous and discrete formulations is straightforwardly established. The minimization operates on a 3D embedded graph whose minimal cut is a solution of the discrete problem, leading to a global minimum. This new approach handling both occlusions and discontinuities has been demonstrated on real sequences, giving remarkably detailed surface geometry up to 1/10th pixel.

@TechReport\{PSQ03,
  author       = "Paris, Sylvain and Sillion, Fran\c{c}ois and Quan, Long",
  title        = "A Volumetric Reconstruction Method from Multiple Calibrated Views using Global Graph Cut Optimization",
  institution  = "INRIA",
  number       = "4843",
  month        = "June",
  year         = "2003",
  keywords     = "Volumetric reconstruction,glob",
  url          = "http://maverick.inrialpes.fr/Publications/2003/PSQ03"
}