To address the first stage, a method of providing automatic calibration for the input image sequence based on a texture planar surface (ground plane) is discussed.

The main body of the talk presents an algorithm for automatically obtaining a segmentation of the object of interest in a sequence of images which was recently published at this year's British Machine Vision Conference. Until recently, the best segmentation results have been obtained by interactive methods that require manual labelling of image regions. The proposed method requires no user input but instead relies on the camera fixating on the object of interest during the sequence. It begins by learning a model of the object's colour, from the image pixels around the fixation points. The algorithm then proceeds by extracting image edges and combines these with the object colour information in a volumetric binary MRF model. The globally optimal segmentation of 3D space is obtained by a graph-cut optimisation. From this segmentation an improved colour model is extracted and the whole process is iterated until convergence. It is observed that this fixation constraint, which requires that the object of interest is more or less central in the image, is enough to determine what to segment and initialise an automatic segmentation process. In addition, it is shown that by performing a single segmentation in 3D, we implicitly exploit a 3D rigidity constraint, expressed as silhouette coherency, which significantly improves silhouette quality over independent 2D segmentations.

The talk concludes with an overview of the multiple view stereo algorithms developed within the University of Cambridge and the Toshiba Research Laboratory to generate the final 3D models.