Overview

In the first upcoming section we will describe the adaptation of the structure and motion recovery algorithm of the AR-System. Although the main goal is the recovery of motion of the camera throughout the video sequence, the system also recovers a crude 3D structure of the real environment. This can be useful to handle problems like resolving occlusions and extracting the illumination distribution of the real environment. We will focus on the motion recovery abilities of the AR-System.

In a following section we will discuss the use of the recovered motion parameters and the 3D structure to register virtual objects within the real environment. This involves using the crude 3D representation of the real environment which we obtain as an extra from the motion recovery algorithm. Dense 3D reconstruction of the real environment is not necessary but may prove useful for future solutions to the occlusion problem.

As input to the AR-System we can take totally uncalibrated video sequences. The video sequences are neither preprocessed nor set up to contain calibration frames or fiducial markers in order to simplify motion and structure recovery. Extra knowledge on calibration parameters of the video camera can be used to help the AR-System to recover motion and structure but is not necessary to obtain good results.

The video sequences are not required to be taken from a purely static environment. As long as the moving parts in the real environment are small in the video sequence the algorithm will still be able to recover motion and structure.