Skip to Main Content
Multiple camera views of a scene are utilized to detect and reconstruct object surfaces in three dimensions. Special attention is paid to the reconstruction of occluded objects which are only partially visible. Input images can be obtained from either an array of cameras or a single moving camera. The formulation is based on a capture and display technique developed in the optics community. Various sharpness metrics are explored for estimating the locations of visible surfaces in the volume, and reconstruction is posed as an optimization problem. Realistic object reconstructions are demonstrated on sets of data consisting of real images.