The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Voi. XXXVII. Part B4. Beijing 2008 
For disaster monitoring from airplanes, a near real time sensor 
and processing system will be developed at DLR. This system 
consists of three digital cameras, an onboard processing unit, a 
microwave data link, and a mobil ground station with 
processing units (see Figure 1). Image data will be distributed 
directly to the security related ground forces in cases of 
disasters. 
The 3K camera system consists of three non-metric off-the- 
shelf cameras (Canon EOS IDs Mark II, 16 MPix). The 
cameras are arranged in a mount with one camera looking in 
nadir direction and two in oblique sideward direction (Error! 
Reference source not found.2), which leads to an increased 
FOV of max 110°/ 31° in across track/flight direction. 
«I 
ISPli 
Figure 2 DLR 3K-camera system consisting of three Canon 
EOS IDs Mark II, integrated in a ZEISS aerial 
camera mount 
The camera system is coupled to a GPS/IMU navigation system, 
which enables the direct georeferencing of the 3K optical 
images. Interior camera parameters were determined by a 
laboratory calibration (Kurz, F., 2007). Error! Reference 
source not found.3 illustrated the image acquisition geometry 
of the DLR 3K-camera system. Based on the use of 50 mm 
Canon lenses, the relation between airplane flight height, 
ground coverage, and pixel size is shown, e.g. the pixel size at a 
flight height of 1000 m above ground is 15 cm and the image 
array covers up 2.8 km in width. 
Coverage 
Figure 3 Illustration of the image acquisition geometry. The 
tilt angle of the sideward looking cameras is approx. 
35°. 
2. DSM GENERATION WITH 3K CAMERA SYSTEM 
2.1 Automatic processing scheme for DSM generation 
An overview of the automatic processing scheme for the 
generation of DSM using 3K imagery is given in this chapter. It 
is deciding that the processing flow works without user 
interaction fully automatic up to the DSM (see Figure 4). 
Input data are three overlapping images (backward, nadir, 
forward) taken from the same camera. The image attitudes and 
positions are measured with a GPS/IMU navigation system (IGI 
AeroControl lid) in the differential mode with accuracies up to 
0.01°/0.1m in the real time case and up to 0.003°/0.05m in the 
post-processing case. Other input data are the geometric 
calibration data separated in parameters of the interior 
orientation for each camera and the boresight alignment angles 
for each camera. The determination of interior orientation 
parameters must be done offline, whereas the estimation of 
boresight angles could be automated and also performed online 
(Kurz, F., 2007). 
The most time consuming part of the DSM generation is the 
surface point matching (92% of the total processing time), 
which is divided into the subpixel hierarchical matching based 
on interest points (Lehner, M. 1992) and the subpixel region 
growing matching (Heipke, C., 1996; Otto, G.P. 1989). 
Figure 4 Automatic processing data flow for generation of 
DSM including calculation time in % and points 
filtering by 1) cross check by additional matching 
and by 2) outlier detection during the forward 
intersection. 
Images must be pre-orientated to find out robust starting points 
for the hierarchical matching. For this, affine transformation 
parameters between the images are derived from GPS image 
positions. The hierarchical matching starts with the “Fòrstner” 
interest operator at the top of the image pyramid in the nadir 
image. Then, a bidirectional matching based on the normalized 
correlation coefficient is performed for every image pair (nadir- 
backward/nadir-forward). The matched points are used for the 
refined definition of search spaces in the next image pyramid 
layer, where the interest operator is performed again and the 
1166