This program implements a full structure from motion pipeline, from detecting and matching image features to computing 3d structure and camera positions.
It's based on Noah Snavely's bundler. The original code has changed quite a bit, mainly by using OpenCV for feature detection, matching and homography/fundamental matrix estimation. Furthermore, Eigen is used for linear algebra, SBA was dropped in favor of Ceres for bundle adjustment. wxWidgets is used for the GUI, glm and GLEW for OpenGL related stuff. You can find a complete list of used libraries under compiling.
Select a folder containing .jpg files (sample images are provided in the Testdata folder) and hit the big play button, et voila. IMPORTANT: The jpg. files must contain EXIF tags with focal length! Depending on your computer's memory and speed, it may be advisable to resize the source images. For reference, on my system (Core i7 quadcore, 16 GB Ram), image sizes up to 3000 x 2000 pixels are tested and work well.
As with many other feature based sfm programs it is recommended to take pictures with wide angle lenses and a sufficient but not too large baseline. Rich textured, uniformely lit and non-reflective surfaces are optimal candidates for reconstruction.
In order to see something in the 3d viewport, a graphics card capable of OpenGL 3.3 and above is needed.
Results will be automatically saved to a .ply file in the directory of the source images. Exporting the solution to MeshLab, Autodesk Maya, .out files (bundler) and CMVS is possible. Blender export is planned for the future.
Building the program from source is not possible yet because my OpenGL framework (glocyte) is not released. This will change soon, so stay tuned.
As for the rest of the dependencies, here is the list:
The program is currently being developed for Windows, a Visual Studio 2013 solution is provided.