— Mw ww 7 
-—- AW ww - 
THE CONSTRUCTION OF DEM'S USING AERIAL VIDEO IMAGE SEQUENCES 
Luuk Spreeuwers and Zweitze Houkes 
Laboratory for Measurement and Instrumentation 
University of Twente 
Netherlands 
luukQmi.el.utwente.nl, hksQmi.el.utwente.nl 
KEY WORDS: Digital Elevation Map, Image Sequence, Remote Sensing 
ABSTRACT 
In this paper a method for construction of Digital Elevation Maps (DEM's) using aerial video image sequences is presented. 
Using video images sequences implies that many images of relatively low resolution are available. From these image sequences 
position measurements of point and line features are obtained and measurements of surface orientation. This results in several 
different types of measurements with different accuracies. A method is presented that can combine these different types of 
measurements and provides a measure of accuracy for each point in the DEM. 
1 INTRODUCTION 
In this paper a new method for fully automatic generation of 
Digital Elevation Maps (DEMs) from video image sequences 
is presented. The video sequences are obtained from an aero- 
plane that is flying along a straight line and a camera looking 
either straight down or with a fixed angle. The viewing direc- 
tion is always perpendicular to the flight direction, see fig.1 
The application of video image sequences means a dense time 
sequence of images is available with, relative to aerial photog- 
raphy, low resolution. The objective is to obtain an accuracy 
of better than 1 m for the height estimation and a ground 
resolution of better than 1 m also. 
EAN aeroplane 
e 
se camera 
e ~ 
  
Figure 1: Illustration of the recording of video image se- 
quences. The aeroplane is flying perpendicular to the pa- 
per and the viewing direction of the camera is perpendicular 
to the flight direction. The camera does not have to look 
straight down, however. 
1.1 Type of scene 
The scenes we are concentrating on are mainly agricultural 
scenes, which means mostly textured scenes with relatively 
few clearly detectable feature points and position measure- 
ments will tend to be less accurate. On the other hand the 
positions of clearly detectable features like corners of roads 
can be determined with high accuracy. Furthermore in agri- 
cultural scenes generally the surface is smooth without steep 
cliffs. Any steep height transition must be an object on top 
of the surface itself. Hence the problem can be formulated as 
the desire to build DEM's from relatively low resolution video 
image sequences for scenes that contain few clearly distin- 
819 
guishable feature points and a rather smooth surface. 
1.2 Requirements 
The proposed method must be able to handle different types 
of measurements with different accuracies. Furthermore an 
attractive feature would be if the DEM could be built se- 
quentially, since the measurements during a flight are also 
obtained sequentially. Closely related to this is the require- 
ment of being able to update or fill in an existing DEM, i.e. 
new measurements can be used to improve the accuracy of 
an existing DEM, while the existing DEM can be used as an 
initial estimate. An important requirement is further that for 
each position the accuracy of the estimation of the elevation 
must be known. 
1.3 Methods 
The approach that is proposed in this paper combines a num- 
ber of techniques to realize this goal. Feature point tracking 
is used to obtain high accuracy height estimates for the few 
available clearly distinguishable feature points. Shape from 
shading techniques are used to obtain estimation of the local 
orientation of the landscape. Edge features are also tracked 
through the image sequence to obtain height measures for the 
borders of fields and other objects. Finally the more classical 
approach by using stereo images can be used to obtain depth 
measurements (see e.g. [Cochran, 1992]). However, these 
will only be accurate if the difference between the viewing 
angles of the stereo image pair is large. Since the difference 
of viewing angle is small for successive images, the images 
will have to be stored for a longer period. The other meth- 
ods do not require this. 
The proposed method for DEM construction consists of two 
processes: 
e propagation of the measurements 
e combination of measurements 
The first process describes what can be deduced about the 
height at a position p2 given a measurement at position pl 
with a certain accuracy and the current DEM. 
The second process describes how a new measurement can 
be used to update the DEM and makes use of the first 
process. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996