Bayesian surface reconstruction

We illustrate an innovative method, which estimates the surfaces -modelled as polygonal meshes- bounding objects present in a scene, viewed by arbitrarily placed cameras. We present a Monte Carlo based iterative approach which, at every step, increases its knowledge about the scene sampling the unknown volume around the current estimation. Then, the samples, which mostly appear to be consistent with the measurements, are used to extend the mesh representing the surface. The reconstruction is regularized applying a filter -based on a dynamic system- to the mesh. This operation will preserve the high curvature areas of the surface, while smoothing away the noise in the estimation.