178
of an integrated CAD system. Multi photo matching (4) is one of the major benefits of digital photogrammetry, espe-
cially if applied in close range. This should be possible, both with gray values as well as with feature points by incorpo-
rating geometric constraints like epipolar lines. Multi photo matching gives the prospect of high measurement preci-
sion and robustness (5) and can overcome problems like occlusions. System autonomy (6) is certainly most desir-
able, since as long as pre-knowledge is required, time consuming interactive input will remain. At least, all process
steps should be independent from initial values. The self diagnosis of the system (7) with respect to point accuracy and
problem areas has to be taken for granted. The system has to provide a reliable quality measure which can be used
later on for editing in the CAD-system or on a digital stereo plotter. Since also non-metric digital cameras like still
video cameras can be used with the system, calibration procedures (8) are to be considered and integrated into the
system. Characterline detection (9) from point clouds has already been addressed. If characterlines are visible as gray
value edges in the digital images (for instance artificially marked with tapes), line following algorithms based on geo-
metrically constrained multi photo matching (Baltsavias, 1991) could be applied for direct measurement of the charac-
terlines. Finally, the integration (10) of the measurement system into a CAD-system would lead to a powerful tool
which could be flexibly applied in the whole design process chain.
Altogether, the keys to the inherent power of the system are dense surface digitization (3) and multiple photo matching
(4). The system plays the role of a fast and robust point cloud generator which provides a dense surface description.
The dense point cloud is coded with reliable quality indicators and can be used in CAD-systems for surface recon-
struction and visualization. Especially, the integration into a modern CAD-system will lead to a turnkey system making
an efficient and fast reverse engineering feasible. In this way it will contribute to new trends in modern car design con-
cepts. The system is currently under development with regard to all items mentioned above, except for characterline
detection (9), which has been postponed, for the time being.
3. MEASUREMENT SYSTEM
The steps to be worked through within the photogrammetric system are the capture and the digitization of the images,
the measurement of control and tie points including a subsequent bundle adjustment, and the surface reconstruction.
If non-metric cameras are used, they have to be calibrated in advance. Figure 1 gives an overview for the characteris-
tic workflow. In the following, the process steps are described in more detail with regard to their mathematical and
technical aspects.
3.1. On-line triangulation Digitization of Digital images from
: analog images CCD camera
The current version of the MATCH-I system is de-
signed to measure control points and tie points inter-
actively, as well as semi-automatically. This means Reconstruction of orientation
that the human operator uses the cursor to provide e semi-automatic measurement of control and
initial positions for the subsequent refinement based tie points
on least squares matching techniques. Multiple image e bundle adjustment
matching is applied, using the classical affine trans-
formation and a consistant parameter setup (Agouris,
1992). In order to promptly get orientation results, the
e calibration
interactive measurement tool incorporates an inte- Surface reconstruction
grated bundle adjustment package which provides e preparation of single project
orientation data without initial values and statistical e batch processing of all projects
analysis of gross errors based on data snooping tech-
niques (see 3.2). This highly efficient on-line triangula-
tion tool gives maximum precision for image coordi- Output in CAD-format
nates and reduces the evaluation time for orientation
parameter.
Post processing
manual editing
3.2 Bundle adjustment
e filtering
Since the bundle adjustment package runs without
initial values for the orientation parameters and un- Figure 1: Workflow of photogrammetric measurement
known object points, the orientation can run fully system
automatically from scratch. Only the results have to be
checked for blunders in the observations; this is supported by data snooping.
3.2.1 Orientation
The orientation is split up into three steps whereby the first two provide approximate values for the final step, which is a
conventional bundle adjustment.
Besides the image coordinates there are some additional observation types as
IAPRS, Vol. 30, Part 5W1, ISPRS Intercommission Workshop "From Pixels to Sequences", Zurich, March 22-24 1995
wb À
fees
(oq
O {+ 3
gb ch
a
