Automatic recovery of camera poses based on satellite image sequence captured by a high frame rate image sensor

Satellite imagery have the advantage of rapid and cyclical monitoring of changes in environment, land use, natural resources, etc. Registering a satellite image sequence and further making it into a video by recovering camera poses will strengthen the merit mentioned above. The images used in this study are captured by a push-broom and frame-based camera mounted on an aircraft flying in high altitude, which is used to simulate a micro-satellite, and the images are captured with high frame rate. Frame-to-frame registration between subsequent frames should be done first. Speeded-Up Robust Features (SURF) is the feature-based image matching algorithm used for feature point detection, description and matching, and followed by Random Sample Consensus (RANSAC) to remove the wrong matching pairs. In order to improve the registration accuracy, overcoming the influence of relief displacement on image registration is needed, that is, only the feature points belonging to the ground point are used for registration. Since the features in the images are not coplanar, there is no simple geometric transformation model to describe the relationship between the corresponding image points, which means that we cannot solve the problem of feature point filtering by analyzing the geometric relationship of the corresponding feature point pairs in image space but should solve in the object space. Therefore, we use the corresponding feature points and the camera matrices obtained by camera calibration to recover camera poses and reconstruct the 3D structure of the scene by triangulation methods. Finally, the feature points are filtered by their altitudes to take out the ground points for precise registration and video production.