The next stage of the project is to automatically compute the image segmentation. This will be achieved by using robust techniques to compute the homographies between the surfaces in the image.
In this paper a new method for computing the projection matrices of an image sequence, has been presented. This approach is an extension of Hartley's algebraic minimisation technique. It is an attractive technique as it requires only twelve parameters to be optimised, regardless of the length of the image sequence. The algorithm has been tested on synthetic data, in the presence of noise. The projection matrices computed in the first half of this paper were then used for visualising three-dimensional models of image sequences. A second application, image editing, was also presented which made use of the calibrated image sequences.
This research is funded by Trinity College Cambridge, the Cambridge Commonwealth Trust and the Overseas Research Students (ORS) Award Scheme.