Finding vehicle boundaries in SAR images

Abstract:
This paper presents a new approach to vehicle delineation in SAR imagery, able to run fully automatically using just the results of earlier stage detection algorithms to provide the initial detections. Recognition of vehicles in SAR imagery is a problem which is likely to require a staged approach in terms of algorithms. These stages may consist, for example, of single pixel detection, filtering, clustering, delineation, discrimination and classification algorithms. In this paper we present a new algorithm for the delineation stage which is concerned with finding the outline of vehicles, thus separating the target response from the image background. The new technique is predicated on finding edges within SAR images and it is shown that a key advantage of this approach is the ability to delineate a single target at a time, avoiding a high-dimensional minimisation problem. Results are compared to an earlier algorithm which gives insight into the impact of different approaches to delineation and the use of different heuristics to perform the fitting.