QUANTITATIVE ASSESSMENT OF THE PROJECTION TRAJECTORY-BASED EPIPOLARITY MODEL AND EPIPOLAR IMAGE RESAMPLING FOR LINEAR-ARRAY SATELLITE IMAGES

Epipolar geometry rectification is one of the critical issues in photogrammetry, which is a strong corresponding searching constraint in dense image matching for 3D reconstruction. In this paper, the properties of the projection trajectory-based epipolarity model are analyzed quantitatively, and the approximate straight line and parallelism property of the epipolar curve are discussed comprehensively using the linear pushbroom satellite images, i.e. IKONOS, GeoEye images. Based on the analysis of the epipolar line properties, a practical method for epipolar resampling developed. In this method, the pixelwise relationship is established between the original and the epipolar images. The experiments on TH-1 images show that quasi rigorous epipolar images can be resampled using our proposed method for both along-track images. After epipolar geometry rectification, the vertical parallaxes at checkpoints can achieve sub-pixel level accuracy, thus demonstrating the correctness and applicability of the proposed method.