96 
Stereo Light Microscope Calibration for 3D Submicron Vision 
Gaudenz Danuser 
Communication Technology Lab, Image Science Group 
Swiss Federal Institute of Technology, ETH 
CH-8092 Zürich 
email: danuser@vision.ee.ethz.ch 
Commision V, Working Group 2 
KEY WORDS: Three-dimensional, Calibration, Bundle, Adjustment, Stereoscopic, Light Microscopy 
ABSTRACT 
The rapid development in micro- and nanotechnology has introduced new challenges to photogrammetry and computer vision. 
Microscopic imagery has to be seriously investigated by these communities. This paper represents a first step in this direction. 
It addresses the calibration of Stereo Light Microscopes. Various new aspects have to be investigated when trying to develop 
computer vision and photogrammetric approaches for this type of sensor. Normally, in macroscopic imagery the contrast and 
image quality can be controlled to some degree. Due to poor contrast and worse quality of the microscopic imagery the reliable 
and precise localization of image coordinates becomes a much more difficult task. A high level vision algorithm followed by 
a modified Least Squares Template Matching adequately solves these problems. In addition, a new imaging model for Stereo 
Light Microscopy is introduced. It is based on the weak perspective situation in microscopy and includes a new distortion 
term describing the non paraxial imaging of a Stereo Light Microscope. The results of various calibration runs demonstrate 
the suitability of this new approach. The new model is explicitly compared with the performance of standard imaging models 
used in computer vision and photogrammetry, when applying them to microscopy. Relative accuracies of laterally 1 : 1000 and 
vertically 2 : 100 are obtained in the object space. In contrast, the relative accuracy in the image space reaches 2 : 10000. 
The discrepancy between the relative accuracies in object and image space results from the limited precision of the underlying 
calibration standard. 
1 INTRODUCTION above. Real-time observation of a 3D scene is impossible 
with all the other microscopes as their data acquisition relies 
on a scanning procedure. Furthermore, with a SLM nearly 
every object, e.g. living cells, can be observed. In addition, 
the low N.A. results in a quite large working distance between 
the specimen and the front lens. This may become important 
when observing objects with a relatively large vertical exten- 
sion or if they have to be handled while viewing at them. 
The development of technologies to handle and manufac- 
ture complex systems on the micro- and nanometer scale is 
a grand challenge for science and engineering. This develop- 
ment is accompanied by a quickly growing demand to pre- 
cisely measure three dimensional micro structures. For many 
applications (quality control, object description, manipulation 
of micro-structures, etc.), it is of utmost importance to gain 
geometrical information from a relatively large field of view. This paper addresses the problem of 3D measurements with 
Thus, a new challenge for the photogrammetry and machine SLM. Studies of the same topic can hardly be found in the 
vision communities is to extend their techniques for extracting literature. To the best of my knowledge, the only relevant 
accurate spatial data to make them applicable to microscope paper has been presented by [Kim et al. 1990]. It de- 
imagery. scribes a method for 3D reconstruction using a SLM. The au- 
thors demonstrate their approach on biological objects. Their 
imaging model is based on an orthographic projection and 
they operate with the viewing angles given in the technical 
specifications of the manufacturer. Neither an initial calibra- 
Quantitative techniques for stereoscopic measurements have 
largely been limited to the Stereo Scanning Electron Micro- 
scope (SSEM). The idea of introducing photogrammetry to 
the SSEM came soon after the development of the first instru- | £1 t 
ments [Maune 1973] and it is still in progress [Gleichmann et tion nor an a posteriori analysis of the 3D accuracy obtained 
al. 1994]. The state of the art in photogrammetric treatment bY their approach has been carried out. They emphasize the 
of SSEM imagery is summarized in [Ghosh 1989]. limitation of their approach and propose that a refined imag- 
ing model, as well as powerful statistical tools, may improve 
the quality of the results. It is exactly the purpose of my 
work to focus on the description of microscopic imaging with 
a photogrammetric approach. 
In contrast to the SSEM, stereoscopic measurements using 
Stereo Light Microscopes (SLM) have not been deeply inves- 
tigated. The limited depth of field in a SLM is most likely 
the key factor that discouraged intensive research in SLM 
imagery. To circumvent this problem, the optical commu- A first paper about this topic has been presented by [Danuser 
nity developed other techniques to reconstruct a 3D micro and Kübler 1995]. A thorough numerical analysis allowed us 
space: confocal microscopy, optical sectioning, laser confocal to define the set up for a novel Bundle Adjustment. Since 
microscopy, etc. [Taylor et al. 1992]. Another difficulty in then, the proposed calibration procedure has been success- 
quantitative stereo light microscopy is to obtain images of fully implemented on a fully operational micro vision system. 
sufficiently high contrast. Specular reflections and specimen Compared to the first paper, this one addresses more the 
which may be transparent further diminish the image qual- practical aspects of microscope calibration. It discusses the 
ity. Above all, the low Numerical Aperture (N.A.) limits the system hardware (section 2) and the key problems of the al- 
sensor resolution to about 2 um. However, SLM provide con- ^ gorithm, namely the image coordinate acquisition (section 4) 
siderable advantages over the microscope types mentioned and the parameter estimation in the Bundle Adjustment (sec- 
101 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B5. Vienna 1996