e The precision is easily controlled with repeated 
measurements. To make this test relevant it 
should span over a longer time span e.g. two 
hours. The reliability is achieved by a high number 
of calibration points. The error tests are only 
looking at the residuals. The self diagnosis are of 
two kinds, internal and external. The internal 
diagnosis uses the fact that more than two cameras 
are used for the intersection of points. The 
external diagnosis is using control points and 
looking at the residuals. 
Technical Description 
2 -22 CCD cameras 
System orientation with known distances, self- 
calibration of system after installation 
Grey level based point measuring 
0.4 points/sec 
The absolute accuracy 0.01 - 0.02 % 
Special details: 
Self-calibration using distances 
Very high accuracy 
Can handle many cameras 
Automatic corrections and self-diagnosis 
5.4 System 4 Track-Eye, Innovativ Vision AB 
a Continuation of earlier 2D system 
b No photogrammetrists in the development of the 
main 2D motion analysis system. For the 3D 
analysis module photogrammetric competence 
were used. 
c The influence of the design of the 3D module. The 
way errors are treated and their effect. 
d Nothing special in the data extraction part. The 
bundle adjustment with self-calibration of the 
unstable part of the interior camera parameters. 
e The high resolution scanner, 6.2 pm, together with 
a grey-level based tracking algorithm ensures high 
precision in the image coordinates. The reliability 
is mainly dependent on the number of cameras. 
The self-diagnosis is fairly well developed with 
residual control of known points which 
automatically starts a new system calibration. 
Technical Description 
2 -6 analogue high speed film cameras 
System orientation with 3D calibration frame, self- 
calibration of un-stable parameters during motion 
sequence. 
Grey level based point measuring/ tracking 
7 points /sec 
The absolute accuracy 10 mm 
Special details: 
Self-calibration connected to self-diagnosis 
| Color monitor 
768 x 512 
Camera TrackEye Software 
Recording 
  
  
  
  
  
Tracking 
Analysis 
  
pr 
Image Work Station 
Video Recorder 
  
  
  
  
  
RE 
[1 
Keyboard Mouse 
ej 
  
  
  
  
  
Es = 
Film Scanner 
  
Videodisc 
fig.11 System Configuration of TrackEye 
6. CONCLUSIONS AND REFLECTIONS 
The main intention of this paper was to recognize 
any differences between photogrammetric real-time 
systems depending on the background of the 
developers. The main characteristics of a real-time 
system can basically be described as: 
Fast and Robust 
The traditional photogrammetric approach, which 
crudely may be described as putting everything into 
large linearized LS problem, may be fine for aerial 
mapping, but the geometrical conditions and time 
constraints in industrial and other close-range 
applications are not always fitted for this. It may be 
described as robust but is not always as fast as wanted. 
The machine vision approach which, very generally, 
may be said to be more attracted by direct solutions, is 
on the other hand fast but not as robust as a correctly 
treaten over-determined system. 
There seem to be a contradiction between these two 
approaches, but it is also possible that a merging of 
the two ideas can be fruitfull. Direct solutions for fast 
estimations of e.g. initial values is engaging many 
researchers which e.g. resulted in the workshop at 
this Congress ,"Calibration and Orientation of 
Cameras in Computer Vision". Similar ideas were 
expressed at the "Second International Workshop on 
Robust Computer Vision" organized by prof. W. 
Forstner in Bonn earlier this year. 
When talking about the terms precision and 
reliability there seems to be a difference in the way 
these are handled. The photogrammetric approach is 
to try to model all errors according to a physical 
model, ending up with many correction terms. The 
other approach is to model the errors independently 
of the sources, by e.g. a matrix with a correction 
vector for each pixel. This latter method is fast and