REGISTRATION OF NEAR REAL-TIME SAR IMAGES BY IMAGE-TO-IMAGE MATCHING

Near real-time SAR images are used at DLRs Center for Satellite Based Crisis Information for mapping natural hazards like flood disasters or landslides. In crisis situations a rapid and precise geocoding is mandatory to produce information within several hours that is comparable with other georeferenced information. Unlikely, the orbit information is often not very precisely known some hours after data acquisition, which leads to misregistration. This paper deals with the development of a fully automatic orientation for SAR images. The goal is to determine a precise orbit for geocoding by image-to-image matching between a near real-time (NRT) image and a reference SAR image. The approach uses a new algorithm for feature extraction developed in computer vision by Lowe, 2004. No prior information about the pose of the images or the overlapping parts is needed. The point operator extracts points with scale- and rotation-invariant descriptors (SIFT-features). These descriptors allow a successful matching of image points even in situations with highly convergent images, i.e. with different incidence angles. Our approach consists of the following steps: Point extraction and matching by using the SIFT parameter descriptors with an extended matching scheme. The resultant points of the reference image are used as ground control points (GCPs) for an adjustment of the SAR imaging geometry of the NRT image. Then, a geocoding of the NRT SAR image can be carried out. This achieves results equivalent to a high precision orbit. Examples of two datasets are presented and the results of the approach are discussed and evaluated. The results show that the approach can be used for a wide range of scenes with different SAR sensors, different incidence angles, and different overlap extensions. The results are very reliable but depend on well structured image content