3D feature estimation for sparse, nonlinear bistatic SAR apertures

We present an algorithm for extracting 3D canonical scattering features observed over sparse, bistatic SAR apertures. The input to the algorithm is a collection of noisy bistatic measurements which are, in general, collected over nonlinear flight paths. The output of the algorithm is a set of canonical scattering features that describe the 3D scene geometry. The algorithm employs a pragmatic approach to initializing feature estimates by first forming a 3D reflectivity reconstruction using sparsity-regularized least squares methods. Regions of high energy are detected in the reconstructions to obtain initial feature estimates. A single canonical feature, corresponding to a geometric shape primitive, is fit to each region via nonlinear optimization of fit error between the complex phase history data and parametric scattering models using a modification of the CLEAN method. Feature extraction results are presented for sparsely-sampled, nonlinear, 3D bistatic scattering prediction data of a simple scene.