The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
839 
working in extreme cases where occlusions exist. These 
occlusion problems are always exhibited during navigation in 
downtown area as resulted from parking or heavy traffic. 
Finally, with linear feature, the reconstruction process is 
practical one as compared to strategies that rely on point feature 
only. 
Of similar importance to the previous features, the proposed 
framework is designed to be flexible when it comes to the 
treatment of the different adjustment objects in the adjustment. 
Two examples are given herein. Firstly, different scientific 
communities use certain parameterization for the rotation 
matrix: navigation community usually expresses the body 
attitude in space using roll, pitch, and azimuth; while the 
photogrammetric community use sequence of Euler rotations. 
Therefore, the framework must allow the parameterization of 
each exposure station independently using any parameterization 
set. Secondly, when dealing with camera interior orientation 
(10), distortions in small format cameras can be modelled using 
first or second order polynomial. Meanwhile, when dealing 
with large format cameras (aerial), distortion models are more 
complicated and need higher order polynomial. Design 
flexibility allows attaching different distortion models to 
different cameras independently. 
7. SIMULATIONS AND RESULTS 
In order to prove the applicability of the proposed fusion 
scheme, several simulations have been performed. The first 
simulation (SI) tests the possibility of using LMMS navigation 
and pictorial data to georeference airborne images based on 
point measurements. The photogrammetric network consists of 
four aerial images (A-l to A-4), and 8 LMMS images taken 
from 4 land exposure stations, with 4 common ground points T1 
to T4. The aerial images are taken with RC30 cameras at 
1000m above ground level, see Figure 4. A minimum of three 
common points are required for datum definition. Other tie 
points between aerial images are still required. 
Figure 4: Photogrammetric Network Configuration 
Table 1 shows statistics of the true errors as estimated from the 
simulated trajectory and their corresponding estimated value for 
airborne positions and attitude; while Table 2 displays the 
standard deviation results. Comparing the results of the true 
errors and the estimated standard deviation, one can reveal that 
the results of the accuracy estimate are pessimistic. 
Max. 
Min. 
Mean 
E(m) 
0.123 
0.012 
0.084 
N (m) 
0.137 
0.011 
0.056 
H (m) 
0.042 
0.002 
0.023 
WO 
20.3 
1.1 
6.8 
PC') 
11.5 
6.1 
9.0 
KD 
8.8 
3.2 
5.9 
Table 1: Actual Absolute Errors Statistics for Point Based 
Scenario 
A-l 
A-2 
A-3 
A-4 
E (m) 
0.088 
0.110 
0.100 
0.090 
N (m) 
0.110 
0.130 
0.130 
0.110 
H (m) 
0.055 
0.074 
0.077 
0.056 
WH 
19.8 
25.2 
25.2 
20.2 
PH 
14.0 
19.4 
18.4 
14.4 
KD 
6.5 
7.2 
7.6 
6.5 
Table 2: Adjustment STDEV 
Previous simulation, using point feature, implements one step 
adjustment for both LMMS and aerial images measurements. 
The application of the previous simulation is limited the 
visibility of point feature in aerial images. To ensure such 
condition, point feature need to be signalized before taking the 
aerial images. This constrain is not realistic, time consuming, 
and sometimes impossible when used with old aerial images 
databases. Moreover, occlusions from moving and parked cars 
may prevent the measuring process. Considering the limitation 
of the previous scenario, another simulation (S2) based on line 
measurements has been performed. The simulation uses the 
same photogrammetric network configuration as in SI. A major 
advantage of this scenario is that conjugate points 
measurements, between the two data sets, are no longer 
required. The linear matching entities simulated from lane line 
marking which present the best alternative due to their ease 
measurement due to high contrast. Table 3 summarizes the 
statistics of the true errors for the aerial images (A-l to A-4). 
Max. 
Min. 
Mean 
E (m) 
0.067 
0.026 
0.045 
N (m) 
0.055 
0.010 
0.030 
H (m) 
0.069 
0.003 
0.034 
W H 
8.8 
0.5 
5.3 
p n 
12.6 
0.1 
8.7 
KD 
16.3 
5.7 
10.5 
Table 3: Actual Absolute Errors Statistics for Line Based 
Scenario 
The selected common lines between the two data sets have to be 
skewed lines. It is obvious from the obtained results that the