A New Stereo Algorithm based on Snakes

As part of the ongoing activities for the European space mission BepiColombo to Mercury, a new stereo-matching algorithm is here proposed: this algorithm uses deformable surfaces, or snakes, to find a dense disparity map. Subject to both external and internal forces, respectively represented by the similarity function and by smoothness constraints on the disparity map, a “deformable” disparity map evolves from an initial approximate state to an optimal one in which the algorithm has reached convergence. This algorithm is expected to provide one of the image matching tools for the Digital Terrain Model generation procedure that will be used by the BepiColombo stereo camera. To check the algorithm, tests have been performed on synthetic images derived from 3D models of geological features relevant to planetary science. The results show that it is possible to obtain an image measurement accuracy comparable to the one attainable with the Least Squares Matching algorithm. In addition, less object smoothing can be obtained since the object points are not derived by a large scale averaging over a terrain patch, as for example, in area-based methods; this means that more details of the terrain can be captured. Finally, because of the continuity constraint, this method is also expected to be robust in case of blunders in the reconstruction of the parallax field.