In this thesis, the modeling of extended objects with low-dimensional representations of their 2D geometry is addressed. The ultimate objective is the classification of the objects using libraries of such compact 2D object models that are much smaller than in the state-of-the-art classification schemes based on (High Range Resolution) HRR data. The considered input information consists of HRR datasets measured at widely separated aspect angles of the object, thus being highly sparse in the angular dimension. Such input datasets are supposedly available from a network of scanning surveillance radars.