tanbul 2004 
  
mu 
cy between 
d discrepancy || 
line H 
crepancy line | 
  
th direction 
* strip). 
ish a fixed 
ever, these 
s due to the 
tween laser 
'ard matter. 
location on 
nates in the 
in order to 
survey was 
1s. A large 
flat football 
is shown in 
racteristics. 
ficant slope 
ng of pure 
les out any 
resence of 
putation of 
| full swath. 
established 
m (DGPS). 
lopographic 
! necessary 
pographical 
€ collected 
iostly at the 
nnis courts. 
IC test arca. 
espondence 
tions and 
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV. Part B2. Istanbul 2004 
consequently the — absolute — planimetric 
determined. 
accuracy was 
5.1 Absolute height accuracy 
A buffer zone (with a width of one meter) was constructed 
around each ground data line and the corresponding laser data 
points inside that buffer zone were identified. Then for each 
laser data point that lies within a meter or less from two ground 
data points, the corresponding height point with the same 
planimetric position was interpolated from the surrounding 
ground-surveyed points. Since the points are near by and the 
surface is flat, a linear interpolation was used. In order to 
minimize the effect of the planimetric uncertainty of the laser 
data, the interpolation was limited to totally flat areas. The 
slope of those areas was restricted to not exceed 1095. Direct 
differencing is then applied to compute the offset between the 
two data points from the laser and ground (the laser height was 
subtracted from the ground interpolated height). More than a 
thousand differences were computed. Figure 7 shows the 
histogram of those differences. The histogram shows a bias in 
the differences of -0.088m with a standard deviation of 
+0.082m. That means that the average height of the laser data 
points included in the test are higher than the surveyed ground 
by 0.088m. The total computed RMSE of the test data from the 
ground reference was about 12cm. This number represents the 
real standard deviation of the LIDAR data heights. 
  
  
Northing * 574000 m 
8 
8 
   
380 400 420 440 ABO 450 500 520 540 
zasling + 913000 m 
Figure 6: Surveyed ground points over the test area and 
Selected locations for the planimetric accuracy computation. 
Counts per bin 
  
de 
a -03 -02 -01 0 oium 02 5 703 
Quantized height diff. in m (Surveyed Height - LIDAR Hieghts) 
Figure 7: Height differences histogram between the laser height 
and the ground height. 
5.2 Absolute planimetric accuracy 
Planimetric offsets are more complicated to determine since 
they require especially significant features to establish the 
correspondence between the two data sets. Such features and 
locations for estimating the offsets may not be available or 
when they exist are usually limited. Moreover, identifying these 
locations is costly in time and requires great care in order to be 
reliable. Drainage ditches, terrain curvature, and building gable 
roofs are some examples of such features. 
Eight locations as shown in figure 6 were identified and 
successfully used in obtaining the planimetric accuracy. Six of 
those locations were used to compute the offset in the X 
(Easting) direction and the other two were utilized to compute 
the offset m the Y (Northing) direction. Offsets in the X 
direction were given more attention since they coincide with 
the scanning direction. In each of these locations the data points 
from both data sets, LIDAR and ground, were identified as 
shown in figure 8(a). From each set, an estimated curve using 
least squares fitting was constructed (green solid line for the 
LIDAR points and red solid line for the ground points) 
representing the available data as shown in figure 8(b). The 
idea here is to match these two curves and obtain the shift that 
will maximize the match. Prior to that, the height bias should be 
removed in order exclude its effect in the matching. In figure 
8(b) the green dashed curve represents the LIDAR data after 
removing the height bias between the two data curves. 
To get the best match between the two curves, the LIDAR data 
curve will be shifted gradually around the ground data set in the 
direction of the computed offset (X as in figure 8). The shift 
ranged from —2m to +2m with an increment of 0.01m. At each 
increment, the offset (in the intended direction X or Y) of each 
ground point and its interpolated-correspondence point from the 
LIDAR curve data is computed. The sum of the squares of 
these offsets is considered as the matching cost at each location. 
Figure 8(c) shows the matching cost function behavior with 
respect to different shift values. At the minimum matching cost, 
which is associated with the best match between the two 
curves, the correspondence shift is obtained. Figure 8(d) shows 
the original data curves, after removing the height bias, and 
alter the planimetric shift.