On modelling three-dimensional objects by uncalibrated cameras

Three efficient techniques are presented for modelling shape of 3-D objects employing uncalibrated cameras. The first technique employing multiple video cameras is proposed for recovering 3-D shape of nonrigid objects by factorization. The second technique employing pairs of facing cameras which surround a rigid object reconstructs an entire shape of the object by collecting all the projected feature points around the object into a single measurement matrix, which is factorized to yield an entire 3-D model of the object. The first and the second technique are combined to yield the third technique which not only models 3-D shape of a non-rigid object at each time but also recovers its deformation process during observation all at once. Satisfactory performance of these techniques are shown by the performed experiments. The present study is expected to contribute to various fields which have interests in 3-D shape/deformation recovery, modelling, and analysis, such as 3-D modelling of characters in video games, 3-D modelling of excavated objects in archaeology, human motion analysis in sports, dancing, or rehabilitation, etc.