3d Structure from Motion from Uncalibrated Cameras

Final year project to build a 3d model of a scene captured with an uncalibrated video camera. The project involves obtaining 3d information from 2d images. This was done by determining a homography between frames in the video sequence and calculating depth information. Several approaches where implemented and compared.



This report addresses the techniques for 3D structure from motion for the case of an uncalibrated camera. The underlying theory is discussed and a number algorithms are compared. A working implementation of all the theory discussed in this report was created and it’s specific details and results are given. It was established that a reconstruction can be obtained up to a relative depth with an uncalibrated camera. Finally a conclusion and suggestions for improvement are proposed.