The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B5. Beijing 2008 
considered as 0.008 degree for heading and 0.005 degrees for 
the RMS of the pitch and roll. Obviously, this performance with 
model 1 can not reach the accuracy required for aerial 
photogrammetry. 
The second test is processed in the model 2 with 12 error states, 
i.e. 3 random constants for the gyroscope drifts, 3 random 
constants for the accelerometer biases, 3 random constants for 
the gyroscope scale factors and 3 random constants for the 
accelerometer scale factors. The innovation and measurement 
residuals are shown in the Figures 11 and 12. The estimated 
errors of position, attitude and gyroscope drift and 
accelerometer bias are shown in Figures 13 - 15. The difference 
of the position between Caravel PP and Throstle is shown in 
Figure 16. The difference of the attitude between the Throstle 
solution and the POSPac solution is shown in Figure 17. The std. 
deviations are about 5m for position, 3cm/s for velocity, 0.002 
degree for pitch and roll, 0.008 degree for heading. 
Figure 11. The innovation of the Filter with model 2. 
Estimation residual of the Filter 
Figure 12. The measurement residual of the Filter with model 2. 
Figure 13. The estimated position error in model 2. 
Figure 14. The estimated gyro drift in model 2. 
Figure 15. The estimated accelerometer bias in model 2. 
Figure 16. The position difference between the results of 
Throstle in model 2 and of GPS. 
Figure 17. The attitude difference between the results of 
Throstlein model 2 and of GPS. 
As shown in Figures 11 and 12, the predicted residual and 
measurement residual of the KF in model 2 are about the 0.1 - 
0.2 meter, which is much better than those in model 1. The 
estimated error of the position in the KF is about 0.02 - 0.05 
centimeters. The estimated drifts of the gyroscopes are similar 
to those in model 1 in general, however, they are much more 
stable in the short time period. The improvement is more 
evident for the biases of the accelerometers, where there is no