1. Introduction

The standard method for photo orientation in photogrammetry is the bundle block adjustment. The photo coordinates of the object points are measured nowadays mainly with high precision automatically by digital photogrammetric workstations. In bundle block adjustments the rays from the object points through the projection centres to the measured photo points represent a spatial pencil of rays for each photo. By means of the bundle adjustment all homologous rays of one point will be optimised to intersect in one point.

If additional parameters and a simultaneous camera calibration are included in the bundle adjustment, the systematic errors in the photographs and in the photogrammetric systems can be eliminated or reduced considerably and improve the precision of the result.

The purpose of adjustment is the determination of the three-dimensional object coordinates of the measured points and the determination of the orientation parameters of the photographs. These orientation parameters are directly available for further use after bundle block adjustment. The adjusted coordinates of object points will be necessary in most cases for further processing of all kinds.

The method of bundle block adjustment can be applied for aerial photos as well as for terrestrial applications. Various survey measurements and photogrammetric observations in addition to the commonly used control points can be included in the adjustment, to strengthen the block.

Normally, the search for data errors plays an important role in adjustment computation. BINGO uses the data-snooping method according to Baarda in an extended version. In the first step detected errors will only be indicated. A further process allows an attended or automatic elimination of the detected errors in photo measurements. Concerning all other types of observations like control points, GPS-data etc., the operator must correct or remove faulty observations himself.

Before adjustment, all initial orientation data and point coordinates have to be estimated. This process includes special error detection methods: the balanced L2-norm adjustment and the RANSAC-method (Random Sample Consensus). This ensures that even in case of several gross errors the initial approximations will not be falsified.