Automatic SAR Target Recognition and Pose Estimation. Part 1. Geometric Methods for Pose Estimation

The paper presents geometrical methodology of determining pose angle of an object against SAR images stored in a data base. The images are separated by a pose angle (less than 2°), obtained from an air surveillance plane at a depression angle and at a distance, producing SAR images of 1 × 1 foot pixel resolution. The data base is public (US Government) for academic and defense research. The data base consists of 100s of commercial and military vehicles, as well as building structures. In this paper, we focus on three targets which have symmetric geometry, each of different size and shape. In addition, we also generate several synthetic targets of various symmetric shapes to serve as the ideal test cases. The analysis is based on simple geometrical considerations as well as edge and surface area determination using number of SAR signatures pixels. The overall methodology aims at significantly reducing data base search space to produce an effective and fast computationally viable search algorithm. The result is very accurate target pose angle determination of the order of 96–98% precision. Once the pose angle is determined, that can be further used to determine target type as well, described in “Automatic SAR Target Recognition and Pose Estimation, Part 2. Statistical Methods for Target Recognition”.