The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
888 
obvious correlation between the biases and the motion trajectory. 
The position difference between Throstle and GPS is about 0.2 
meter after the stability of the filter, which indicates the 
absolute position accuracy in model 2 is about 0.2 meter. The 
differences of attitude angles are less than 0.01 degrees for both 
of pitch and roll, 0.05 degrees for heading. And these values can 
be cut down, especially for the heading, to half, i.e. 0.02-0.03 
degree by smoothing or backward filtering. In this case, the 
Throstle can be considered as accurate as POSPac. Therefore, 
POSPac can not be used as a reference standard to evaluate the 
absolute attitude accuracy any more. Other data, e.g. the attitude 
from bundle adjustment or the coordinates of the ground points 
by traditional surveying can be used to check the final direct 
referencing accuracy. But these methods need the calibration of 
the camera boresight, which is not finished for this test data. So 
in this paper, these two methods are not implemented to 
evaluate the attitude accuracy. While the absolute accuracy 
specification is not achieved, from the analysis above, the result 
of model 2 shows obvious improvement to that of model 1. 
In the 15-state error model a Markov process in the gyroscope 
drift is added to model 2. The std. deviations for position, 
velocity and attitude are almost the same as those in Model 2. In 
this section, the attention is paid to the gyro drift and 
accelerometer bias as shown in Figures 18 and 19, and the 
difference of position (and attitude) between Throstle and GPS 
(POSPac) is shown in Figure 20 and 21. By comparing the 
Figures 18-21 and Figures 14 - 17, it can be found the 
differences are not obvious, which means the improvement of 
the 15-state error model relative to the 12-state error model is 
limited. This is probably due to the limited observation 
capability of the gyroscope data or the unsuitable configuration 
of the stochastic model parameters. The testing and analysis are 
ongoing efforts for which more detailed analysis and results will 
be provided later. 
Figure 18. The estimated gyro drift in model 3. 
Figure 19. The estimated accelerometer bias in model 3. 
Figure 20. The position difference between the results of 
Throstle in model 3 and of GPS. 
The attitude difference between Throstle and POSPac 
Figure 21. The attitude difference between the results of 
Throstlein model 3 and of GPS. 
4. CONCLUSIONS 
In this paper, the 6-state, 12-state and 15-state inertial sensor 
error models are implemented, and the KF performance of each 
is compared and analyzed. The accuracies of the integrated 
system reach 5cm for position, 3cm/s for velocity, 0.002 degree 
for pitch and roll, 0.008 degree for heading in the 12-state 
model, which are all better than those in the 6-state error model. 
However, the improvement of the 15-state error model from 12- 
state error model is limited and insignificant. Further 
investigation is going for the absolute accuracy validation of the 
GPS/INS integration based on different models. And more 
precise stochastic error model will also be tested. 
ACKNOWLEDGEMENTS 
This research is supported by the Chinese National Scientific 
Foundation (No. 40501060). The test flight discussed here is 
supported by the Xinjiang Surveying and mapping Bureau. The 
support of these agencies is gratefully acknowledged. 
REFERENCES 
Abdullah, Q., 1997, Evaluation of GPS-Inertial Navigation 
System for airborne photogrammetry: ASPRS/MAPPS Softcopy 
Conference, Arlington, 1997. 
Feng S., 1999. Research on the low cost 1MU/GPS integrated 
navigation system. Ph.D. Dissertation. Najing University of 
Aeronautics and Astronautics. 1999. 
Grejner-Brzezinska D. A., 1999. Direct Exterior Orientation of 
Airborne Imagery with GPS/INS System: Performance Analysis, 
Navigation, Vol. 46, No. 4, pp. 261-270.