A three-dimensional reconstruction system using space carving

Creating true 3D models of objects in our real world using their 2D images is an area that researchers have started to explore. The traditional approach has been stereo reconstruction. Several techniques to achieve stereo reconstruction have been elaborated. While the current techniques work well under some controlled conditions, they do not provide a reliable solution to object reconstruction under general conditions.;In this thesis, a system that can be used to reconstruct the 3D shape of an unspecified and totally arbitrarily-shaped object is presented. The system acquires calibrated images of an object from different views. Using the images captured at arbitrary viewpoints and the calibration data, the system creates a 3D model of the object. Since stereo techniques do not allow accurate reconstruction under general conditions, we depart from the stereo approach to employ a recent technique called space carving [1]. In this technique, we represent the volume as a discrete collection of small cubes called voxels. The reconstructed 3D model consists of the set of voxels that are consistent with a same color on all 2D projections. Because of the long processing time that the existing space carving technique requires, we develop a parallel version of space carving that considerably speeds up the reconstruction. At the end of the reconstruction, a cloud of 3D points representing the final shape is obtained. Using OOGL (Object Oriented Graphics Library) formats, this cloud of 3D points is converted to a model that can be displayed using a 3D graphics software. Accuracy of the reconstruction is judged by inspecting visually the 3D model created.