Hans-Gerd Maas 
  
technique performing matching between more than two strips, e.g. in the case of laserscanner blocks with overlapping 
and crossing strips. 
2.3 Data testing 
This basic formulation of least squares matching with data in a TIN structure has to be extended in a number of aspects 
due to the nature of laserscanner data: 
e Exclusion of points 
Data points on vegetation or other objects with an irregular shape in combination with the characteristics of laser 
pulse penetration, the sampling pattern and the viewing direction may lead to outliers in matching. These outliers will 
deteriorate the quality of the match and may falsify the result of matching. Therefore they have to be removed by 
robust estimation techniques. If a prior segmentation of the dataset has been performed, all points on vegetation 
should be excluded. 
e Planar triangles 
A severe problem is caused by occlusions typically occurring at one side of buildings close to strip boundaries as a 
consequence of the central perspective of laser scanning across flight direction. These occlusions become visible as 
gaps in one of the patches. Due to the composition of laser- 
scanner data blocks of parallel strips, these occlusions will 
usually have points (on the ground and possibly also on the 
wall) in the corresponding patch (Figure 2). 
As a consequence of this fact and of the minimisation of the 
sum of height differences between the original data points of 
both patches and their interpolated corresponding values, the 
solution of LSM will be shifted by half the width of the occlu- 
sion area. In the case of symmetry, including a whole building 
into the patch will compensate the effect, but in general such 
occlusions may cause biases of several decimeters in the 
solution of LSM. Obviously, the same problem occurs when 
interpolating data points to a regular grid and applying raster- 
based LSM. 
  
  
Figure 2: Occlusion at building gable 
To avoid these effects, points falling into TIN-meshes with irregular size or shape are excluded from the matching 
process. Triangles in occlusion regions can be recognised by their longish shape as well as their steepness, which is 
defined by the geometry of the occlusion as well as the triangle's position in the strip and the flight parameters. This 
option performs well for large occlusions such as occlusions behind gables, but may be ineffective for occlusions 
caused by smaller objects. See chapter 4.1 for further refinement of the technique. 
2.4 Matching strategy 
A suitable matching strategy depends on the goal of the application. For the main application field of the developed tool 
in the strip adjustment of airborne laserscanner data, a matching strategy should provide regions well distributed over 
the strip overlap area, which do allow for the determination of all three shift parameters. 
So far, manually chosen patch locations have been used for controlled testing of the method. Interest operators as used 
in many photogrammetric applications did not show fully satisfactory results yet. As buildings proved to be suitable 
objects in many cases, techniques for the segmentation of laserscanner data and the detection of buildings (e.g. Maas 
1999a, Oude Elberink and Maas 2000) and techniques for modelling buildings (e.g. Maas 1999b, Maas/Vosselman 1999) 
form a viable alternative for indicating suitable regions for matching. 
3. First results 
To evaluate the potential of the technique, it was applied to a the laser scanning test block 'Eelde'. The dataset 'Eelde' 
contains a small city in the Netherlands. It was flown with a 10kHz Optech system at a flying height of 500m with a strip 
width of 250m and a point density of 0.3 points per square meter, corresponding to an average point spacing of 1.8 meter. 
The dataset consists of 10 strips flown in north-south direction and 10 strips flown in east-west direction. Both first and 
last pulse echoes were recorded; only the last pulse data were used for LSM in order to reduce effects caused by 
vegetation. 
  
550 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 
As af 
in the 
12m, : 
e Set 
e Th 
cm. 
e Th 
wa 
poi 
spc 
par 
13 
In pai 
result 
with t 
probk 
vicini 
larly « 
A mo 
result 
succe: 
X/Y/z 
4. R 
A ger 
fulfill 
deterr 
direct 
contri 
showi 
In the 
neigh 
in Fig 
These 
regioi 
singu 
perpe 
This | 
variat 
comb 
stand: 
In pra 
ters w