GEOMETRIC ANALYSIS OF 3D OBJECT POSITIONING 1USING SAR AND OPTICAL IMAGES

摘要

Synthetic Aperture Radar (SAR) and optical images are two major sources in environment remote sensing. The integration of these two datasets can help us to obtain more useful information. From geometric point of view, there two types of data may be combined for 3D positioning. Orientation modeling for satellite images is an important task for 3D positioning. To link an image point with its counterpart on the ground, Rational Function Model (RFM) has advantages of standardization for satellite image processing and is easy to implement. Thus, we use RFM to integrate SAR and optical sensor orientation data for 3D positioning. There are four steps in this study: (1) building RFM for images, (2) virtual point generation with simulated errors, (3) 3D object positioning, and (4) validation. Most high-resolution optical satellite companies provide the imagery with RPCs instead of the ephemeris data, but SAR satellite companies do by contraries. Thus, the generation of RPCs for RFM starts from radar back projection in the first step. Then, using RPCs to build up the RFM may integrate two sensor imagers. We simulate error-free virtual points in the overlap area and add errors with normal distribution on the simulated observations to evaluate positioning errors. For a pair of conjugate points in SAR and optical images, we may formulate four equations to determine the 3D object coordinates. In the study, we have the test data including one COSMO-SkyMed image and SPOT images with different tilt angles. Experiment results indicate that the highest accuracy may be achieved when the convergent angle is smallest. On the other hand, when the convergent angle is closing to 90°, the positioning error is getting large or even diverges. These error characteristics fit with geometric rules.