In this work, we present an adaptive space carving method for scene reconstruction from a set of images obtained from low cost calibrated webcams. Our method uses a combination of silhouette and photometric information to efficiently carve the shape of the observed scene out of a volumetric space represented by an octree data structure. In this method different resolutions are considered both in object space and in image space. This led us to adopt a strategy in which the information used by the photo-consistency test is registered in scene space by projective texture mapping. Another important question addressed in this work is the high level of noise present in low cost webcams. To deal with this problem we devised a statistical photo-consistency test that uses statistical estimators for the noise introduced by the sensors of the cameras.