An Alternative Approach to the Triangulation of Spot Imagery

Satellite line scanner imagery such as that of SPOT and MOMS-02 is usually analysed via a collinearity equation approach based on projective transformation. This paper discusses the application of an alternative approach to the triangulation of satellite line scanner imagery, namely affine projection. The approach involves an initial transformation of the original perspective line image into an affine projection, which is then followed by a linear transformation from image to object space for stereo geometries (cross-track and along-track). The triangulation phase for the affine image data is in turn in two stages. In the first a preliminary adjustment is performed to determine initial values of the six affine transformation parameters for the selected ‘orientation image lines’. The interval of orientation images is about 2,000 lines and the behaviour of the transformation parameters between adjacent orientation images is assumed to be linear. This phase of the process utilises ground control points only. A bundle type triangulation adjustment then follows, in which affine transformation parameters are refined along with the object point triangulation. Initially, the paper reviews features of the affine projection model with special reference being given to the perspective-toaffine conversion for each image line. The method for acquiring initial values for the bundle adjustment is also discussed, along with ground control considerations. Application of the affine orientation model to a SPOT scene covering the Kobe/Osaka area of Japan is then reported, for both Level-1 and Level-2 cross-track imagery. The results of the triangulation indicate that ground check point accuracies at the level of 6m in planimetry and 7.5m in height can be obtained, for both Level-1 and Level-2 imagery, for ground control configurations of 6 or more well distributed points. The results obtained are very encouraging, especially for the geometrically corrected Level-2 imagery. They illustrate that, in spite of some minor theoretical shortcomings with the formulation of the affine projection model, ground point triangulation accuracies equivalent to and in cases better than alternative approaches based on perspective transformation models can be achieved.