An automatic image based modeling system by patch growing

In this paper, we present an automatic and efficient image based modeling system which can create objects' 3D models directly from images captured from different viewpoints. The system firstly uses structure from motion to generate camera parameters and sparse 3D patches. Then, a conservative plane based sweep stereo method on GPU is used to compute quasi-dense depth maps which usually have many holes and create many new patches. Finally, a simplified patch based surface growing method is used to compute dense depth maps together with a dense 3D geometry shape model. Different from others, our system constructs patches from reconstructed points by structure from motion. These patches grow denser gradually from sparse, quasi dense to dense through the following expansion steps. During expansion, conservative strategies are used and object normal is not optimized but gradually computed from reconstructed quasi dense neighbors. This makes the system work better and faster on more difficult scenarios, e.g. wide baseline, varying illumination. Experiments show that our method can generate high fidelity 3D object shape models quite efficiently.