Corner extraction from radar images

The 3D structure of a moving object such as a satellite may be estimated from a sequence of radar images using machine intelligence techniques. A region growing approach for the extraction of features such as the corner of a panel is explored. The grown regions are seeded by strong radar returns. Since the extracted regions are fragmented and do not lead to an outline as would be perceived by a human observer, an adaptive nonlinear algorithm was designed to obtain an improved estimate of the outlines. Corners were detected by a method based on the polygon approximation of the outlines. The proposed corner detector was compared to a number of other corner detectors and was found to be superior.<
>