DEPTH MAP RECONSTRUCTION USING GRAPH CUTS ALGORITHM FOR 3D MODELING

ABSTRACT

Depth map reconstruction is the most method in 3D modeling from 2D images, because depth map has depth information from the images. There are so many researchs about reconstruction images into 3D in different methods. This thesis developed a 3D modeling software using energy-based graph cuts algorithm that reconstructs depth map from six images, and models it in 3D space.
Three major step in this thesis are camera calibration, depth map reconstruction, and 3D modeling. Camera calibration method used to estimate intrinsic and extrinsic camera parameters that used to get correlation points in the depth map reconstruction process. Depth map reconstruction process was comparing the depth configuration that estimated before, with new depth value. Comparing process in depth map reconstruction solved by energy minimizing with graph cuts algorithm. The energy function that used was data and smoothness energy. 3D modeling method that used was triangle mesh and it implemented with OpenGL.
The result showed that the depth map which reconstructed is not good enough, it has 43.3% average error and 36.99 MSE. The average execution time in all conditions is about 310.03 detik using images of 320x240 pixels in 2GHz Intel Core2Duo processor and 2GB of RAM.


You can download :
Journal (PDF)
Full text (PDF)
Sourcecode (VS .NET 2008 / C++)
Research Data