The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
Although the coplanarity method is more robust against
blunders, it takes more processing time. As a final note, when
using control planar patches, one should make sure to use
planar patches with varying slope and orientation in order to de
correlate the estimated parameters in the bundle adjustment
procedure.
METHOD
Surveyed GCP
Coplanarity
Method
(Patches)
Weight Restriction -
Object Space
(Patches)
Coplanarity
Method
(Lines)
Weight Restriction -
Image Space
(Lines)
Meanxx (nt)
0.05
1.09
1.05
1.00
1.01
Meanly (nt)
-0.01
-0.70
-0.75
-0.76
-0.78
MeanAz (nt)
0.14
0.53
0.58
0.65
0.68
CTx(m)
0.09
0.11
0.14
0.10
0.11
cr Y (nt)
0.11
0.12
0.14
0.10
0.10
<y 7 (m)
0.25
0.10
0.10
0.11
0.11
RMSE\ (nt)
0.10
1.09
1.06
1.00
1.02
RMSEy (nt)
0.10
0.71
0.76
0.77
0.78
RMSE Z (nt)
0.28
0.54
0.58
0.66
0.68
RMSE Tom! (nt)
0.32
1.41
1.43
1.43
1.45
Table 1: Mean, standard deviation, and RMSE analysis of the 24 Check Points using surveyed GCP, LiDAR areal control features,
and LiDAR linear control features
3.1.3 Georeferencing Results Using Linear Features:
The results from the georeferencing of the imagery using
LiDAR-derived linear features are presented in the last two
columns in Table 1. It can be noted that the bias between the
LiDAR reference frame and the GPS coordinate system
detected in Section 3.1.2 is also visible in the results of the
experiments using linear features. This is seen in the relatively
large amount of bias in the results (Mean^x, Meanly, Mean¿¡r).
However, the standard deviations (<j x , <Jy, cr z ) are reasonable,
and are compatible with the results of the experiments done
using areal features. That is, the horizontal standard deviation is
similar to the results from experiments conducted using GCPs,
while the vertical standard deviation is improved compared the
results obtained using GCPs as the control features. A possible
reason for this, as suggested in Section 3.1.2, is that many more
linear control features were used in comparison to the number
of ground control points (50 linear control features versus 24
ground control points). That is, the improved vertical accuracy
may be due to the higher redundancy. This bias value has
affected the final values of the root mean square error (RMSE X ,
RMSEy, RMSE Z , RMSE Tolal ).
3.2 Qualitative Analysis
Three orthoimages were generated using the angle-based true
orthoimage generation methodology, developed by Habib et al.
(2007). They were generated using the perspective image
shown in Figure 5a, a digital surface model, and the three sets
of EOPs resulting from using GCPs, LiDAR patches with
weight restriction (in object space), and LiDAR lines with
weight restriction (in object space) as sources of control.
Figures 5b, 5c, and 5d illustrate the differences between the
generated orthoimages using the EOP obtained using GCPs,
LiDAR patches, and LiDAR lines, respectively. By examining
these orthoimages, it is clear that the generated orthoimage
using LiDAR patches and the generated orthoimage using
LiDAR lines are compatible (Figures 5c and 5d). This matches
with the quantitative analysis in the previous sections where it
was seen that indirect georeferencing using either areal or linear
LiDAR control features gives comparable results. Moreover,
the orthoimages generated using LiDAR patches or lines appear
to be more accurate than the orthoimage generated using GCPs.
This can be observed in the orthophotos, where there are more
traces of building boundaries in the latter orthoimage (Figure
5b). Therefore, the EOP generated using GCPs were less
accurate than the EOP generated using LiDAR patches or lines.
Figure 5: a) Perspective image, and orthoimage using a) GCPs,
b) LiDAR patches, and c) LiDAR lines, as the source of control.
4. CONCLUSION
The availability of LiDAR data allows for alternative sources of
control data in photogrammetric indirect georeferencing. In this
regard, LiDAR-derived areal or linear control features can be