Together with the unknown ARI process paramters aı we obtain a
highly nonlinear estimation problem which can be solved in an
iterative computation scheme (Lindenberger 1987). The condi-
tions to be fulfilled by the data for successfull solution of
the GM model and the VCE are mentioned in the same paper (e.g.
the variances on 2 and ce? have to be of same order of magni-
tude).
Gross errors in the observations would disturb the validity of
the GM model in eq.(2). These gross errors are automatically
detected and are taken into consideration by individual weights
in eq(2). The weights are calculated following the robust esti-
mation theory (Danish method, Krarup et al. 1980). On the other
hand, a robust treatment of eq. (3) reduces the influences of
edges and discontinuities in the data, which disturb the ARI
model.
2.3 Capability of the algorithm
What are the main results from the algorithm ? Under the
assumption that the true trajectory of the sensor platform can
be modelled by an ARI-process, we obtain a filtered data set x:
which is the most probable representation of the true track.
Together with the estimated ARI process parameters, several
demands of further evaluation of the sensor data will be
satisfied.
In addition, the ARI-model yields important results, relevant
for the analysis of the stochastic model. The estimated vari-
ance of the observation noise On 2 describes the observation
process. The variance of the prediction errors oe? gives a
fidelity measure how well the the ARI-model is suited for the
real physical process. The inversion of the normal equation
system out of equations (2) and (3) provides accuracy criteria
of the filtered data set, especially the variance of the fil-
tered data ox 2 and the autocorrelation coefficients r(h). It is
emphasised here that all stochastic results are obtained
without any a priori information.
Any systematic effects in the time-series, such as drifts of
the orientation parameters with time, cannot be taken into
consideration by the algorithm. For this reason the estimated
accuracies must be understood not as absolute but as relative
accuracies.
3. Applications
3.1 Position coordinates from NAVSTAR-GPS
The NAVSTAR Global Positioning System GPS enables the determi-
nation of the x,y,z position coordinates of one or more GPS
receivers. The application of GPS is of particular interest in
photogrammetry for the inflight positioning of the aerial
camera. This reduces drastically the ground control require-
ments for aerial triangulation (Friep 1986).
In the case of a stationary GPS receiver the accuracy estima-
tion of GPS measurements is relatively simple due to the redun-
dancy of the observations. In contrast to this, the application
84