A procedure for 3D model construction from sparsely and irregularly sa
mpled points is presented. A two-stage dynamic deformation process is
presented which is designed to produce desirable mesh properties despi
te difficult data characteristics. In a first phase, a mesh of springs
is snapped down to the convex hull of the data. In the second phase,
a pseudo-gravity model is used to attract the mesh points into concave
surface patches. This modeling technique is a new contribution to dyn
amic modeling methods. This process reduces the undesirable effects of
oversmoothness, local concentration, and folding that result from the
sparsity and randomness of sampled data. Our experiments show that th
e proposed deformation process preserves to some extent both the shape
and size uniformities of the patches constituting models. Furthermore
, our modeling process fits surfaces with prominent concavities withou
t prior segmentation of input data. (C) 1996 Academic Press, Inc.