George Vosselman 
  
3 
o 
Pas Ph d 
wh 
  
  
Figure 3. Unfiltered (left) and filtered (right) laser data with a point density of 5.6 points/m2 shown in a grid of 2 m (top) 
and 0.5 m (bottom). The filter were results obtained with the maximum filter. All points higher than 8 m are imaged 
white. The contrast in the lower right image is stretched. The height difference at the ditches in this image is only 0.5 m. 
5 RESULTS 
After filtering the original data with the maximum filter, the resulting DEM showed a detailed structure of ditches 
below the vegetation (figure 3). In the left upper image, some vegetation points can be found just above the dike. This 
arca was under water. Since there arc no measurements on the water surface, the filter selected the lowest points of the 
vegetation. In the enlarged part, small errors due to low vegetation can be observed. The brighter spots inside the 
ditches most likely are points on shrubs. Thus, it has to be kept in mind that these filter results, that are used as a 
reference for the results on datasets with a lower point density, are certainly not perfect. 
  
  
  
  
  
  
  
  
  
Thc percentages of type I and II errors in the Point density Type ! Type II Total 
results of filtering the datasets with a reduced (points/m°) max : prob | max : prob | max : prob 
resolution are given in table l. As to be 1 00 : 23 S0 20 22:3 7253 
expected, the number of type I errors is 1/4 00 141 87 5 $2 40 . 30 
smaller for the maximum filter than for the 1/16 18 ' 3] 139 ! 101 73 | 63 
probabilistic filter. Since the maximum filter 
allows larger height differences, the number Table 1. Percentages of type I and II errors 
  
940 International Archives of Photogrammetry and Remote Sensing. Vol, XXXIII, Part B3. Amsterdam 2000.