The facility identifies image coordinates that significantly (in a
statistical sense) violate the mathematical model of the bundle adjustment. These images
are automatically excluded from the solution, marked and reported in the resulting (.LST)
file. The error detection is formulated and executed according to the following scheme:
- Blunders are defined as those statistically significant deviations in image
measurements which are based on the initial estimates of camera stations parameters
(positions and attitudes). Blunder edit is performed when Visual Basic is executed in the
Intersection Mode with automatic edit option selected. When such deviations are detected,
they are reported in full together with their effect on related camera stations and object
points. The error report produced by Visual Giant must be examined carefully. Reported
errors must be verified and removed before proceeding with the triangulation adjustment.
Blunder detection may need to be performed several times before realizing a blunder free
block of photographs. It is strongly recommended that the blunder detection be completely
satisfied before proceeding with the aerotriangulation process.
- Automatic error editing is the process by which random errors, which are
statistically significant, is identified and removed from the adjustment solution. Visual
Giant performs this task by automatically performing a sequence of complete adjustments
cycles. The upper limit for the number of these adjustment cycles is a user input (see
figure ???). After each one of these adjustment cycles, all image residuals are compared
to their statistically expected values. If the comparison failed at the 90% significant
level, the corresponding measurements are excluded from the block and another complete
adjustment cycle is performed. This process continues until one of the following two
conditions is satisfied :
- No new image measurements are excluded or previously excluded image measurements
are reintroduced in the solution.
- The number of requested error detection cycles has been exhausted.
|