MEASUREMENT OF INDUSTRIAL SHEETMETAL PARTS WITH
CAD-DESIGNED DATA AND NON-METRIC IMAGE SEQUENCE
Yongjun Zhang *^
a. School of Remote Sensing and Information Engineering, Wuhan University, Wuhan, 430079, China
b. College of Hydropower & Information Engineering, Huazhong University of Science & Technology, China
yongjun_zhang@sina.com, yjzhang@supresoft.com.cn
Commission V
KEY WORDS: Photogrammetry, Industry, CAD, Image sequence, Matching, Reconstruction, Measurement
ABSTRACT:
A novel approach for three-dimensional reconstruction and measurement of industrial parts with CAD-designed data and non-metric
image sequence is proposed. The purpose of our approach is to automatically reconstruct and thus measure the producing
imprecision or deformations of industrial parts mainly composed of line segments and circles with information extracted from
imagery. Non-metric image sequence and CAD-designed data are used as sources of information. Principles of 2D and 1D least
squares template matching to extract precise lines and points are presented. Hybrid point-line photogrammetry is adopted to get
accurate wire frame model of industrial parts. Circles, arcs and lines connected to each other on the part are reconstructed with direct
object space solution according to known camera parameters. The reconstructed CAD model can be used for visual measurement.
Experimental results of several parts are very satisfying, which shows that the proposed approach has a promising potential in
automatic 3D reconstruction and measurement of widely existed industrial parts mainly composed of lines, circles, connected arcs
and lines.
1. INTRODUCTION
Three-dimensional (3D) reconstruction from imagery is one of
the most active areas in Close-Range Photogrammetry and
Computer Vision. Since Computer Aided Design (CAD) is
widely used and most industrial parts have their corresponding
CAD-designed data, precision evaluating and quality control of
industrial parts with reference of CAD data receive much
concern in industrial communities. Reducing manpower,
maintaining high precision and consistency, and time of
measurement are the main foci of researchers.
Coordinate Measurement Machine (CMM) is the mostly used
measuring equipment in industrial communities. But the cost
and speed of CMM are still major problems to be resolved
(Steven, 1999). Stereo vision technique with two CCD cameras
and two infrared LED lamps is used by (Kosmopoulos 2001) in
measurement of gaps on the automobile production line, results
of about 0.1 mm precision within an area of 80mm X 80mm are
obtained. Along with the development of computer vision, two-
dimensional automated visual inspection has been widely used
in Printed Circuit Board product lines (Moganti et al 1996).
Although automated vision metrology getting more and more
mature (c.f. Fraser 1999, Pappa et al 2002), there is no practical
3D vision system that can substitute man efforts in imprecision
measurement and quality control for widely existed small-scale
industrial parts such as sheetmetal parts.
Line photogrammetry is used in the reconstruction of objects
mainly represented by polyhedral models. Several approaches
have been proposed to reconstruct architectures in the last years
(Debevec 1996, Heuvel 1999). Photogrammetric techniques are
also used in 3D reconstruction of industrial installations
(Vosselman 2000) and reverse engineering (Ermes 2000) with
CAD models. Although all of these systems are semi-automatic
and time-consuming, line photogrammetry still shows a well
potential in automatic 3D reconstruction especially in the case
of known initial model of the interested object.
In this paper, a new approach of reconstructing and measuring
industrial sheetmetal parts with CAD-designed data and non-
metric image sequence is proposed. The general strategy is to
reconstruct and measure industrial parts quickly and accurately
using techniques of hybrid point-line photogrammetry and
direct object space solution with CAD-designed data and
information extracted from the imagery. À planar grid is used to
calibrate the non-metric CCD- camera and provide initial values
of camera parameters during reconstruction. CAD-designed
data represents the initial model of the part and the topology of
the reconstructed one. Least-Squares Template Matching
(LSTM) and several matching results are discussed in section 2.
Afterwards, detailed approach of reconstructing industrial parts
mainly composed of line segments, circles, connected arcs and
circles is presented. The reconstructed CAD model can be used
to measure producing imprecision or deformations. System
overview and experimental results are discussed in section 4.
And section 5 concludes the paper.
2. LEAST SQUARES TEMPLATE MATCHING
2.1 Line Template Matching
Image points and lines are the most effective characteristics
used for 3D reconstruction in photogrammetry and computer
vision. If *Minimization of the squared sum of grey-difference
Mem ni V" am eeu
d.
