ranbul 2004 
AUTOMATING THE CHECKING AND CORRECTING OF DEMs WITHOUT 
REFERENCE DATA. 
D. Skarlatos *, A. Georgopoulos 
NTUA, School of Surveying, Laboratory of Photogrammtry, 9 Iroon Polytexniou, Athens 15780, Greece, 
(dskarlat,drag)@central.ntua.gr 
Commission ITI 
KEY WORDS: Automation, correction, quality, orthorectification, DEM/DTM, reference data, algorithms, geometric. 
ABSTRACT: 
The need for Digital Elevation Model (DEM) checking has raised since the automated matching techniques emerged. The use of 
reference data is a luxury and therefore cannot be used in all cases. This article presents a completely new mathematical model, 
which translates discrepancies between two orthophotographs created from different photographs, into precise corrections of the 
DEM. These corrections are the differences from the real surface and, if applied over the existing DEM, can produce a more accurate 
one. The mathematical model is straightforward, does not approximate, and therefore there is no need for iterations. 
In order to test the algorithm a reference DEM has been manually collected and distorted with a known pattern. The corrections 
produced by the algorithm follow the known pattern. Tests over automatically created DEMs by commercial software has also been 
made and compared against the reference DTM. Obvious applications include checking of automatically created DEMs, refinement 
of existing DEMs using aerial photographs and update of orthophotographs based on the previous DEM and new imagery. 
1l. INTRODUCTION 
Digital Elevation Models (DEMs) are currently the 
bottleneck of the photogrammetric workflow. Automation of 
aerial triangulation (using GPS, INS and proper software) and 
orthophotograph creation (automatic mosaicing) has stressed 
the problem. The need for orthophotomaps, which are 
becoming a standard and therefore the need for DEMs, city 
models and telecommunication industry expansion are 
stressing the demand of large updated and accurate DEMs. 
On the other hand close range projects for the production of 
point clouds around objects are becoming more and more 
attractive to a wide range of customers. 
Manual collection of points is laborious and extremely slow. 
Digital stereoplotters are becoming cheaper than the past, but 
are still expensive hence usually operators work in shifts. 
Automated matching methods seem to be a solution, but 
proved valid only for small scales, and there is always the 
need for corrections, or at least visual checking, which means 
valuable streoplotter time (ISM, 1998). A completely 
automated method, which could work unsupervised 24 hours 
à day, in a simple computer is very attractive. So the main 
problem is how to check and correct the automatically 
created DEM, without human intervention. The same 
problem also holds for large national organisations, which 
must check subcontractors and delivered DEMs. 
A new matching algorithm is being developed in the 
Laboratory of Photogrammetry in National Technical 
University of Athens (NTUA). During its last stages, where 
tuning and final adjustments are necessary, the urge of 
checking the results upon different objects in different scales 
became evident. Manual collection of a reference DEM is the 
most reliable and obvious solution. for comparison, but if a 
number of models are under investigation then it becomes 
impractical and time consuming. 
A simple solution for checking could be the use of internal 
Statistics, which provide a measure of precision but not a 
measure of accuracy (Zhilin, 1993a; Zhilin, 1988), was 
rejected. 
  
* 
553 
Simple overlay of the two orthophotographs and subtraction 
of the grey level values provides a coarse measure for spatial 
distribution of errors, but not their exact magnitude. 
Therefore this was also rejected. 
Norvelle (1994; 1996) has introduced Interative Orthophoto 
Refinment (IOR), a method where the discrepancies between 
two orthophotographs created with the same DEM but from 
different aerial photographs, were translated in height 
displacement and used to correct the initial DEM. Although 
theoretically the orthophotograph should be independent 
from the initial photograph, in practice orthophotographs 
created from different photographs differ slightly. The 
mathematical model of the corrections was simple and 
approximate. Height correction was calculated using the 
approximate formula: 
H 
dh 2 dx — (1) 
B 
where dh = height correction 
dx = the x difference (in ground units) 
between orthophotographs created [rom the 
left and the right photographs of a pair 
B = base 
H - flying height. 
Although the formula was approximate, a couple iterations 
produced promising results. Since 1996, there wasn't any 
other report on this subject that the authors are aware of. The 
idea of using the discrepancies between two 
orthophotographs to correct the underlying DEM has a strong 
geometric background and seemed attractive to the authórs, 
who decided to investigate further and work out a precise 
mathematical model for the height error in any given position 
using orthophotographs created from left and right 
photographs of a pair (from now on referred freely as left and 
right orthophotographs). 
Corresponding author. This is useful to know for communication with the appropriate person in cases with more than one author.