An original scheme to automatically generate digital terrain models, DTM's, and orthoimages from SPOT stereopairs in a unified way is presented. In addition to modeling the time-dependent orientation parameters, we developed an algorithm to generate the epipolar stereomate incorporating an initial DTM. The refining of the DTM is then accomplished by performing least squares templet matching along each conjugate epipolar line pair, space intersection, and grid value interpolation. Recursively, the epipolar stereomate is regenerated according to the refined DTM. The procedure is repeated until the disparities between the two stereomate images are small enough. The DTM that we generated illustrates the ''apparent'' surface. It is more correct to use this ''apparent'' DTM than the one depicting the ''ground'' surface for the geometric correction for remotely sensed data. Both images of the final epipolar stereomate are actually orthoimages because they are geometrically identical and georeferenced. Experimental results indicate that the orthoimages reach an accuracy up to 2/3 pixels. The rms of the disparities between the two orthoimages, which describes the geometrical identification of the images when the DTM is revised, is less than 1/3 pixels in each of the two test cases.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING