Stitching images of dual-cameras onboard satellite

Abstract

The way of installing dual-cameras on one satellite is adopted to further enlarge the imaging swath, thereby improving the efficiency of data capturing. In this case, stitching images of dual-cameras with high precision is a key step in the practical application. Due to the inadequate overlapping area of dual-cameras, stitching their images by classic methods may cause internal accuracy loss of the mosaic image. The reason is that classic methods estimate the geometric transformation of dual-cameras merely by a few unevenly distributed precise tie points in overlapping area of dual-cameras, which is similar to the case of using unevenly distributed ground control points (GCPs) in block adjustment. This paper proposed a new method to precisely stitch images of dual cameras without losing internal accuracy. First, a model was built to recover the relative geometric relation of dual-cameras and eliminate Charge-Coupled Device (CCD) distortions of each camera, then a virtual camera model depending on the calibrated geometric relation was adopted to achieve a seamless mosaic image. The panchromatic images of camera A and camera B onboard Yaogan-24 were collected as the experimental data. Experiment results show that the calibration accuracies of dual-cameras are better than 0.3 pixels, and the stitching accuracies can reach the sub-pixel level, ranging from 0.3 to 0.5 pixels. On the other hand, the positioning accuracies with GCPs of the mosaic image and of individual camera are better than 0.6 pixels and 0.5 pixels respectively, so the internal accuracy loss of the mosaic image only reaches 0.1 pixels, which can be neglected. This demonstrates that the proposed method can achieve seamless mosaic images without losing internal accuracy.