337 
THEORY AND ALGORITHMS OF DSM GENERATION FROM MULTI-LINE-ARRAY 
IMAGES MATCHING 
LeiRong*, Fan Dazhao, Ji Song, Zhai Huiqin 
Zhengzhou Institute of Surveying and Mapping, Zhengzhou, 450052, China - leirong@163.com 
Commission VI, WG VI/5 
KEY WORDS: Multi-line-array digital images, DSM; CCD; ADS40; Correlative Coefficient; POS 
ABSTRACT: 
A new multiple images matching theory model is proposed in this paper to generate DSM from aerial three line array digital images. 
Through this model, multiple images (more than 2) can be matched simultaneously and the epipolar line constraint can be used 
indirectly. Theoretically, the occlusion and multiple solution problems, which are unavoidable in traditionally image matching 
process, can be greatly solved by this model. Based on the robust multi-image matching algorithms, intelligent DSM generation 
procedure is constructed, and some key techniques during the DSM generation process are investigated. Experiments prove that the 
DSM generation methods proposed in this paper can effectively generate reliable DSM form multi-line-array digital images.. At the 
end of this paper, problems remaining to be studied are presented. 
1. INTRODUCTION 
With the development of sensors and their application 
technique since 1990s, it will not take a long time for film 
cameras to be replaced by CCD digital cameras., which is one 
of the most significant developing trends of aerial 
photogrammetry. Now there are two developing trends for CCD 
digital camera, one is big plane array and another is linear array. 
Currently, plane array CCD doesn’t have enough pixels to meet 
the requirements of practical photogrammetric applications. To 
solve this problem, one way is combining a few plane array 
CCD to produce a big one, but it is expensive and brings the 
other problems in the real-time transferring and storing of 
massive changing data. Additionally, some arrays of plane array 
CCD are randomly distributed, which will result in the losing of 
image pixels and the bringing of more parameters for geometric 
and radiometric correction than linear array CCD. Thereby, 
under the existing research condition, linear array CCD digital 
cameras are the optimal choice for aerial photogrammetry. At 
present, some leading experimental and commercial digital 
camera such as WAOSS, MOMS, WAAC, DPA, ADS40, TLS 
and JAS and so on are all three line array CCD digital cameras. 
Among these digital camera systems, ADS40 is the first 
commercial airborne three line array digital camera system 
developed by Leica Company and the DLR institution. 
Digital Surface Model (DSM) is often referred to as the model 
for the first reflective or visible surface. DSM is an vital 
product of digital photogrammetry and plays an irreplaceable 
role in such aspects as determining objects’ height, generating 
DEM, producing true ortho-image, automatic recognizing and 
extracting buildings and so on. Recently, the techniques of 
digital sensor have undergone a significant development and 
many sensor systems can obtain multiple images of the same 
area simultaneously. As an example, for ADS40, there are 3 to 
7 highly overlapping images on the same flight strip (the 
overlapping degree of neighboring images is nearly 90%) for 
any imaging area. Moreover, the ADS40’s overlapping degree 
of neighboring flight strip is around 60% and this provides 
more images for the imaging area. The high overlapping images 
provides redundant information for automatic DSM generation. 
However, it is quite complex to generate DSM form three line 
array digital images and multiple image matching technique is 
one bottleneck. To generate dense and precise DSM, such 
problems as image occlusion, multiple solution, noise, surface 
discontinuity and so on have to be solved effectively, which 
requires new image processing techniques. 
2. MULTI-IMAGE MATCHING ALGORITHM MODEL 
As is shown in Figure 1, consider three ADS40 digital images, 
which are obtained on the same flight line. 10, II and 12 are 
nadir, forward and backward image respectively, where 10 is 
selected as reference image, and Ii (i=l, 2) are selected as 
searching images.The basic working process of AMMGC 
model can be described as follows. 
(1) Define or extract points pi(i=0, 1, 2, 3, ...) to be matched on 
the reference image. 
(2) For each point pi, determine its approximate height Zi and 
height error AZi. Zi and AZi can be predefined by users, or 
obtained by rough matching of high layer image pyramids or 
initial DSM. The quasi-epipolar lines of pi on the searching 
images are determined by known exterior parameters and line 
fitting methods. Define an image window W around pi, and W 
is named as correlation window. 
(3) On one searching image’s quasi-epipolar line, select a 
searching window (the same size as the correlation window) 
around each point of the line, compute the correlation 
coefficient between the searching window and correlation 
window, and find the local max correlation coefficients. 
Corresponding author. This is useful to know for communication with the appropriate person in cases with more than one author.