METHODS OF THE BUNDLE BLOCK ADJUSTMENT OF PLANETARY IMAGE DATA 
   
W. Zhang, B. Giese, J. Oberst, R. Jaumann 
DLR, Institute of Planetary Exploration, Germany 
Commission III/ / Working group 1 
KEY WORDS: planetary image data, Sequent method, Robust, Baarda, variance estimation, bundle block adjustment 
ABSTRACT 
We have developed procedures for bundle block adjustments of planetary images obtained by framing cameras. The "Sequent" 
method was developed for the elimination of a priori gross errors in the input image point coordinates during the preparation of the 
data. The "Robust" and the "Baarda" methods eliminate any further errors during the block adjustment. The procedure also identifies 
and eliminates systematic offsets and drift in navigation observations. Finally, variances of image coordinates, navigation parameters 
and object coordinates of control points are estimated. The procedure was verified using simulated data and finally applied to lunar 
Clementine image data of a region in the eastern Mare Orientale. The precision of the object coordinates was improved from a 
standard deviation of 2500 m before the adjustment, to 150 m after the adjustment. 
1. INTRODUCTION 
The bundle block adjustment plays a central role in the 
photogrammetric analysis of image data. The procedure allows 
us to correct nominal navigation parameters, and yields precise 
object coordinates of conjugate image points. Adjusted 
navigation data will typically be used in the production of 
DTMs (Digital Terrain Models), mosaic images, orthophoto 
images, and control point networks. In comparison with the 
classical aerial photogrammetric techniques, the bundle block 
adjustment of planetary images faces several challenges: 
navigation parameters are only known to a limited precision, 
and they can contain systematic errors, such as offset and drift. 
Furthermore, on planetary surfaces, normally very few control 
points are available. This paper expands on these specific 
problems and presents existing or develops new methods for 
their solution *). 
Finally, the methods are applied to Clementine images that were 
obtained in two areas on the Moon near Mare Orientale. The 
adjusted results show that these methods are robust and reliable. 
2. CHARACTERIZATION OF PLANETARY IMAGE 
DATA 
The position parameters of a spacecraft, and the orientation 
parameters of a framing camera are inertially measured and are 
known to a certain precision. These quantities can be 
transformed into six navigation parameters (X, Y, Z, 9, 0, 
K) in the photogrammetric system which are introduced into 
the bundle block adjustment as uncorrelated observations. 
Images are typically obtained from several different orbits, or 
even from different cameras in different missions. Therefore, 
  
*) The methods described in this paper are implemented in the 
software package "DLRADJUST" which was written by the 
first author (W.Z.) and has been installed on computers at the 
DLR Institute of Planetary Exploration, Berlin-Adlershof. 
navigation parameters of single images can also have different 
precisions. 
Typically, for most planetary surfaces, there are very few 
control points, and their precision is usually worse than the 
pixel resolution of the surface seen in the image. Even for the 
Moon there are less than 2000 control points (Davies et al., 
1994). 
3. PREPARATION OF IMAGE DATA 
First, large numbers of image coordinates of conjugate image 
points are measured manually on the computer screen. These 
are transformed into the photogrammetric system. 
For the bundle block adjustment, initial values of object 
coordinates of these conjugate points are needed. These are 
calculated using the forward intersection method for each point. 
For the elimination of gross errors in the image measurements, a 
procedure has been developed, we will refer to as the "Sequent" 
method. Initially the best stereo image pair is specified. This 
pair must satisfy simultaneously three requirements, a long 
base, a small standard deviation of conjugate points, and a small 
standard deviation on the unit weight. If the standard deviation 
of the conjugate point, for all pairs, is larger than the given limit 
value, then this point is regarded as a "bad" point and is 
eliminated. If the best pair for this conjugate point is found, we 
continue the adjustment for this conjugate point. In every 
subsequent step, a new image point is introduced. If the 
standard deviation of this conjugate point is larger than the limit 
value, this new image point is designated as a "bad" image 
point, and is eliminated, thus removing all large errors. From 
this we obtain a reliable set of object coordinates for the future 
bundle block adjustment. 
4 METHOD OF THE ADJUSTMENT 
In the second stage of the process, the actual bundle block 
adjustment is performed. The navigation parameters and object 
coordinates are the unknowns. The image coordinates are from 
1004 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B3. Vienna 1996