A new tool for image understanding is stated in this paper: the visual hull of an object. Many algorithms for identifying or reconstructing 3-D objects use the 2-D silhouette of the object. If a non-convex object S is considered, some features of the surface of S can be useless for identification based on silhouettes; the same features of S cannot be reconstructed by volume intersection algorithms using a set of silhouettes extracted from multiple views of the object. Broadly speaking, we define the visual hull of an object or set of objects S as the envelope of all the possible visual rays tangent to S. Only the features of the surface of S which lies also on the surface of the visual hull can be reconstructed or identified using silhouette-based algorithms. After a suitable general discussion, this paper presents an algorithm for computing the visual hull in 2-D. A precise statement of the visual hull concept appears to be new, as well as the problem of its computation.
KeywordsConvex Hull Multiple View Tangency Point Fourier Descriptor Incidence Lattice
Unable to display preview. Download preview PDF.
- M. Hebert and T. Kanade, “The 3-D profile method for object recognition”, Proc. IEEE Comput. Soc. Conf. on Computer Vision and Pattern Recognition, pp.458-463, June 1985.Google Scholar
- S.A. Dudani et al. “Aircraft identification by moment invariants”, IEEE Trans.Comput., Vol. 26, pp. 39–46, 1981.Google Scholar
- S. Baase, “Computer algorithms”, Addison-Wesley, New York, 1988.Google Scholar
- G. Salmon, “Analytic geometry of three dimensions”, Cambridge, England: Metcalfe, 1874.Google Scholar
- J. Stewman and K. Bowyer, “Constructing the perspective projection aspect graph of non-convex polyhedra”, IEEE Intern. Conf. on Comput. Vision, pp. 494-500, 1988.Google Scholar