Accuracy Analysis of Generalized Imaging Models for QuickBird Stereo Imagery

Abstract

This paper analyzes the accuracies of different generalized imaging models (but RFM) for a separate-orbit QuickBird stereo imager pair, including extended DLT model, parallel projection model, projection-modified affine model, parallel perspective model, time-variant-parameterized affine model and the proposed time-variant-parameterized parallel perspective model. The experimental results have revealed the spatial dynamic characteristic of the QuickBird sensor. Due to the continuous re-orientation of the sensor, imaging models with constant parameters such as extended DLT model and affine projection model, can not describe the imaging geometry of the QuickBird imagery sufficiently. By contrast, imaging models with time-variant parameters perform much better. Meanwhile, the parallel perspective model with time-variant parameters shows superiority to the time-variant-parameterized affine model. With no less than 14 ground control points, parallel perspective model with time-variant parameters can provide the accuracy of sub-pixel level, which is close to that achieved by the more rigorous RFM.