Thesis (M.Sc.) - Cairo University - Faculty of Engineering - Department of Computer Engineering

3D models creation is one of the most attractive areas in computer vision. Robotics, gaming, simulation, virtual museums and preserving world heritage are just few examples of applications in which 3D models can be used. One of the most challenging tasks in this area is building the 3D model using set of images taken for a specific scene or object. In most practical cases, the software that intends to build the 3D model has no information about the camera movements or its internal parameters like focal length which can be all random and change from one image to another. Computer vision algorithms are applied to recover cameras projection matrices that holds information about both cameras relative movements and its internal parameters. The computational complexity of most common 3D modeling algorithm which is incremental structure from motion (SfM) is known to be of O (n⁴). In a long sequence of images extracted from video with couple of hundreds or thousands of images the computational cost of incremental SfM becomes impractical and subject to many problems

Issued also as CD