Charles Toth 
  
4.2 Sampling Pattern 
The footprint of the laser beam and the ground pixel size of the EarthData LIDAR and digital camera system as 
discussed above are very comparable and also both systems work with similar FOV’s. For our discussion, the typical 
LIDAR data acquisition configuration described above will be considered. Fig. 3 depicts a surface patch showing the 
LIDAR spots and the back-projected image (stereo pair) pixel tessellation. As illustrated, there are three independent 
irregular sampling patterns. Of course, the irregularity depends primarily on surface undulations and to a lesser extent 
on the sensor orientation. The ratio between the LIDAR and image samples is about 1:60. Since the LIDAR system may 
receive multiple returns, the effective sampling size for this rare situation can be smaller as indicated on the lower-right 
LIDAR footprint in Fig. 3 where the laser beam hits a break line. A completely missing LIDAR spot is another likely 
anomaly; for example, due to surface slant or due to specific materials such as tar (which has no response in the narrow 
LIDAR spectral band) it is possible that no laser return will be detected at all. 
     
  
: Image 2 pixel tessellation 
+ 
pun 
() LIDAR footprint 
Figure 3. Footprint distribution of the image ground pixels and LIDAR spots 
To assess the impact of the different sampling rates of the LIDAR and imaging sensors with regard to the surface 
extraction problem, two approaches can be considered. First, if the sampling rate of the LIDAR system (typically 
defined by the cross track direction) is adequate to properly describe the surface, i.e. Eq. 4 is satisfied. This is usually 
the case for rural areas with modest surface undulations. In these situations, the use of image data to support the surface 
extraction process is rather limited and is mainly restricted to fill in areas with missing LIDAR spots. Except for these 
rare cases, the primary purpose of the simultaneously acquired image data is visual coverage, the ortho-rectified 
backdrop of the surface. The second and more important case is when the sampling rate of the LIDAR system is not 
adequate for the required surface representation with respect to the requirements of the mapping objective. This is the 
typical case for urban areas and will be discussed in the following. 
4.3 Under-sampling over Urban Areas 
Surveying of densely built-up urban areas is in high demand and yet this is one of the most difficult mapping tasks to 
perform. This is primarily due to the large number of man-made objects with lots of vertical surfaces, occlusions. 
shadows, moving objects, etc. Probably the surface discontinuities, generally called break lines, represent the most 
difficult problem, and from a strictly theoretical point of view, they would require a diminishingly small sampling 
distance. Consequently, this is the case where anything that can increase the sampling frequency for the LIDAR system 
is appreciated. Multiple laser returns, which are used primarily for vegetation separation, can virtually increase 
sampling rate locally by providing two (or more) observations for one laser pulse; for example, from the ground and 
from rooftops. However, this is a very rare scenario since the probability that the laser beam hits the edge of a building 
is very small. Therefore, the only way to introduce additional information into the surface extraction process is the use 
of simultaneously acquired imagery. It is important to note that the images come fully oriented. On the one hand, digital 
cameras capture them and thus the interior orientation is automatically given (basically preserved from the camera 
calibration). On the other hand, the LIDAR system assumes the use of a high-quality direct sensor orientation system, 
which by design easily provides the exterior orientation data. 
  
902 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000.