ACCURACY ANALYSIS OF LARGE SIZE DIGITAL AERIAL CAMERAS

摘要

A test area north of Philadelphia was flown with Vexcel Imaging UltraCamD, UltraCamX and Z/I Imaging DMC digital frame cameras as well as with the 3D-CCD-line scanner camera Leica ADS40 and the analogue RC30. The frame images have 60% overlap in both directions. The object resolution is approximately the same with ～5cm ground sampling distance (GSD). 42 well defined control and check points are available with sufficient accuracy. Even if the nominal GSD is approximately the same, this must not correspond to the effective object resolution. This has been investigated by edge analysis -a sudden change of the grey values in the object space is causing a continuous change of the grey values in the image. A differentiation of the grey value profile leads to the point spread function and from the point spread function the effective resolution can be determined. Corresponding to the information about the modulation transfer function in the calibration report, UltraCam-images have in the corners a lower modulation transfer function. In the case of pan-sharpened UltraCam images in general a loss of up to 30% of the image quality against original panchromatic images has been seen. The aerial image, scanned with 13 microns, has a factor for the effective resolution of approximately 1.4; that means it corresponds to an image scanned with 18 microns pixel size. The large size digital frame images are merged from 4 separate panchromatic cameras and the colour cameras. The image deformation of the sub-cameras, determined by laboratory calibration, is respected by the generation of the homogenous virtual images. So by theory they should not show any systematic image errors. In reality an analysis of the image residuals of block adjustments shows very clear systematic image errors corresponding to the merge of the sub-images. The Hannover bundle block adjustment program BLUH has been updated by special additional parameters fitting to the geometric problems of the individual cameras. So by self calibration the exact shape of the systematic image errors -the difference between the mathematical model of perspective geometry and the real image geometry -has been determined. The root mean square discrepancy at independent check points is not so much dependent upon the chosen additional parameters; also with the standard parameters of BLUH approximately the same accuracy could be reached. This is not possible with the Ebner set of additional parameters, here the Gruen set with 44 parameters is required, but such a high number of additional unknowns may cause a loss of accuracy. The block adjustment with the DMC-images is resulting in a sigmaO of approximately 3.5u.m (1/3 pixels), while for the UltraCam it is in the range of 4u.m. For the DMC this is larger like usual, but it can be explained by the very large scale. The accuracy achieved at independent check points with+/-2cm for X and Y is in the range of the given information about the check point accuracy. For this good result it was necessary to check the exact pointing of every object point with the available field image of the control and check points. With ADS40 images a horizontal accuracy of 2 -3cm and a vertical accuracy in the range of 3cm has been reached.