International Conference on Computer Vision, 1998. India.
This paper addresses the problem of automatically constructing triangular spline models of complex objects from a few photographs. Our approach combines silhouettes from registered images to construct a G1-continuous spline approximation of an object with unknown topology. We then use the same optimization procedure to estimate the pose of a modeled object from a single image, and the distance residual is used as a basis for a simple object recognition strategy. The proposed approach has been implemented, and examples of model construction and pose estimation are presented for several complex objects.