Passive capture compared with active sensing exhibits some significant advantages in terms of cost, practicality, complexity of use and integration to a larger system. These are the characteristics that make cameras the more attractive sensor for a wide market reconstruction product. Unfortunately unless accurate calibration and precise information on the pose of the stereo cameras are known, 3D measurements can only recovered for a sparse set of features. As these features correspond to discontinuities in the intensity of the images and are usually associated to ``significant'' structure in the scene they can be reliably matched and tracked along successive frames.
Reconstructing a 3D model of a real environment involves both estimating its geometry and topology. Scene geometry refers to the 3D position of the recovered structure primitives while scene topology refers to the set of surfaces that connect these primitives together. Although significant work has been done towards the first direction, only limited attempts has been made to recover topology.