INVESTIGATION OF 3D GEOPOSITIONING AND DEM ACCURACY OF CARTOSAT-1 STEREO IMAGERY

摘要

Because Cartosat-1 data is one of the ideal sources for topographic map updating, it is important to provide enough geometric accuracy with minimum number of Ground Control Points (GCP). For a 1:25.000 scale topographic map planimetric accuracy requirement is approximately 5 m. and vertical accuracy is approximately is 3 m. In order to maintain these accuracies, geometric orientation better than these values should be achieved. So far many studies have been conducted in 3D positioning matters of Cartosat-1 data. But in a sense, they were somewhat experimental and in a well controlled test site. Here we, as an end user sense with a real study site, conducted our test. Cartosat-1 provides Rational Polynomial Coefficients (RPC) with the raw imagery. It is possible to improve RPCs by a direct or indirect approach. Direct methods update the original RPCs themselves. Indirect approach brings concluding transformations in object space and does not change original RPCs directly. Improvement is achieved by using GCPs with affine transformation or most simple means. In this paper, three different models (polynomial model: 0th order with translation to x and y coordinates, 1st order affine and 2nd order) are used. These different order polynomial models were used in a Cartosat-1 stereo pair covering southern part of Ankara with the different GCP numbers and distributions. GCP and Check Points (CPs) were obtained from an orthophoto mosaic comprising 16K. scale air photos with planimetric accuracy better than 1 m. In each step, some of the GCPs are used as CPs to test the absolute accuracy. Also, we produced a DEM from Cartosat-1 stereo pair and we used a DEM produced from 10 m. interval contours obtained from 1:25.000 scale topographic maps as a reference in order to compare the DEMs,. One and half pixel accuracy with 0th order polynomial, nearly one pixel accuracy with 1st and 2nd order polynomials was achieved. Without GCPs the planimetric accuracy was nearly 500 m. The number of the GCPs does not have significant effect on 3D positioning accuracy after 8 GCPs. For a good and reliable solution lsl degree and minimum 8 GCPs are recommended. By autocorrelation method in Leica Photogrammetry Suite (LPS) 9.1, 10 m. grid interval DTM was created. This DTM was subtracted from 1:25.000 scale contour DTM. The accuracy for automatically created DTM without editing is nearly 4 m. This shows that Cartosat-1 has a good potential for generation of DSMs with a grid spacing of about 10 m and accuracy (RMS) of about 4 m. This is a good approximation for 1:25.000 scale topographic maps. Orthophotos can be produced directly from this DTM. Also contours may be generated after some editing processes. In the light of above test, it can be inferred that by using evenly distributed half pixel accurate GCPs, Cartosat-1 imagery meets the geometric accuracy requirements of 1:25.000 scale topographic maps. Then the remaining activity is a little bit more field work than the aerial photo compilation.