d.
bd
Vy —
%
vr
Guehring, Jens
DATA PROCESSING AND CALIBRATION OF A CROSS-PATTERN STRIPE PROJECTOR
Jens GÜHRING, Claus BRENNER, Jan BÖHM, Dieter FRITSCH
Institute for Photogrammetry (ifp), Stuttgart University, Germany
Geschwister-Scholl-Strasse 24D, D-70174 Stuttgart
Jens.Guehring@ifp.uni-stuttgart.de
KEY WORDS: Calibration, Computer Vision, Accuracy, Performance Analysis, Sensors
ABSTRACT
Dense surface acquisition is one of the most challenging tasks for optical 3-D measurement systems in applications such
as inspection of industrial parts, reverse engineering, digitization of virtual reality models and robot guidance.
In order to achieve high accuracy we need good hardware equipment, as well as sophisticated data processing methods.
Since measurements should be feasible under real world conditions, accuracy, reliability and consistency are a major
issue.
Based on the experiences with a previous system, a detailed analysis of the performance was carried out, leading to a
new hardware setup. On the software side, we improved our calibration procedure and replaced the phase shift
technique previously used by a new processing scheme which we call line shift processing.
This paper describes our new approach. Results are presented and compared to results derived from the previous
system.
1 INTRODUCTION
The industrial manufacturing process has changed over the years. Driven by the need for higher productivity,
development cycles became much faster and time-to-market is more important than ever before. Feature based and
parametric CAD systems allow rapid changes, simulation methods help to guarantee technical soundness and new
technologies such as rapid prototyping are used to establish new production processes. In the context of rapid product
development, quality control becomes a crucial and time-critical factor in development as well as in the production
process.
Traditionally, coordinate measurement machines (CMMs) are used for mechanical part inspection. CMMs are well
established and widely accepted in industry, but suffer from limitations such as high cost and low measurement speed,
corresponding to a long validation time and therefore do not meet the requirements formulated above.
On the other hand, optical 3-D sensors measure the shape of objects, without the need to physically probe surfaces.
Modern optical sensors are faster, cheaper and provide a higher measurement density than conventional techniques and
are therefore ideally suited for applications like reverse engineering, rapid validation (including soft or deformable
surfaces), digitization of VR models and guidance for industrial robots.
After some years of skepticism, optical measurement systems are starting to replace the touch-trigger probes which
have been widely used in industry to date. The performance of such a system depends both on the type and number of
sensors and on the configuration of the entire system. The processing steps needed to convert collected image data to
three-dimensional coordinates play another important role. However, system calibration is without doubt the limiting
factor for the accuracy of most 3-D measurement systems.
In Brenner et al. (1999), we reported on the photogrammetric calibration of an active optical triangulation sensor and
compared the results to a direct calibration method, namely polynomial depth calibration. Since then, we have enhanced
our hardware setup, automated most steps of the calibration procedure and developed a new method to solve the
correspondence problem with sub-pixel accuracy.
The remainder of this paper is organized as follows. We first describe our hardware setup in Section 2. Section 3 details
the patterns and the processing steps we use to solve the correspondence problem. The issue of calibration is addressed
in Section 4. In Section 5 experimental results are given and compared to our previous measurement system. Section 6
summarizes the results we have obtained.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B5. Amsterdam 2000. 327
