Automatic shape reconstruction of rigid 3-D objects from multiple calibrated images

We present a method for reconstructing the shape of a rigid 3-D object based on multiple calibrated images from an arbitrary camera setup. The algorithm is robust and able to reconstruct concavities. It is based on the idea to reconstruct a photo-consistent shape by carving and coloring voxels from a volume description of the object. Coloring and carving steps are carried out with respect to a color verification from the input images for each visible voxel. As compared to other approaches, the threshold used for the color verification is adapted automatically. Information included in the input images is processed and applied simultaneously without any explicit integration or assignment to the volume. The approach is evaluated using camera images of a real scene.