Bundle adjustment
The term comes from the bundle Photogrammetry and refers to optimizing the " Sehstrahlenbündel " a 3D scene that is recorded by several cameras or of a camera from several perspectives. In the bundle at the same time the positions of points in 3D space, the positions and orientations of the observing cameras and their internal calibration parameters can be so adapted to the measurement images that remaining error (eg, image distortion, measurement error of the evaluation) optimally as possible to all observations are distributed.
Specifically, the term is used to evaluate not only individual photogrammetric image pairs (2 each covering measurement images ), but to link any number of contiguous images (block ) with each other.
It does not matter whether such an image block is a contiguous area of the earth's surface ( aerial photography ), or any terrestrial from several points of view captured object of close-range photogrammetry.
One speaks of a bundle, since each pixel of an image measurement defines a beam through the center of projection (camera lens) and all pixels in an image are therefore measured laced in the center of projection into a bundle (central projection).
The linking of the images is achieved by means of corresponding points, which are seen in a plurality of overlapping images, and each can be localized exactly. This procedure is called aerial triangulation (also: image triangulation ), with the help of which an area without ground control points can be bridged. The selection of these points can be carried out or assisted by automated pattern recognition algorithms.
For the calculation using an equation system, which images the optical path described above mathematically. Since considerably more observations ( measurements ) are present as unknown, the errors of the measured image coordinates are minimized with the help of adjustment.
As a result of the adjustment computation is obtained at the same time the unknown orientation elements ( external and internal orientation ) of all windows and also the previously unknown object coordinates of all tie points. Go The more of these points in as many different images in the bundle, the more accurate the subsequent adjustment computation.
Following a successful bundle you can continue with the conventional photogrammetric evaluations and collect any additional objects in three dimensions.