THESIS
2014
xiii, 92 pages : illustrations (some color) ; 30 cm
Abstract
Plastic parts with specular surfaces are common and produced in a large amount. The
measurement of specular surfaces by a machine vision system is the key to surface
quality assurance in the automated production. In the measurement of specular surfaces,
an important issue is the depth ambiguity problem. The previous researches on the depth
ambiguity issue are comprehensively studied in this thesis. Based on Petz’s two screens
method, a new approach is proposed to remove the ambiguity by an extra beam splitter.
The proposed approach inherits the simplicity of Petz’s method and avoids the laborious
and time-consuming task of moving the screen by using a mechanical system.
In order to achieve an accurate measurement, another two important problems are studied
in this thesis, namel...[
Read more ]
Plastic parts with specular surfaces are common and produced in a large amount. The
measurement of specular surfaces by a machine vision system is the key to surface
quality assurance in the automated production. In the measurement of specular surfaces,
an important issue is the depth ambiguity problem. The previous researches on the depth
ambiguity issue are comprehensively studied in this thesis. Based on Petz’s two screens
method, a new approach is proposed to remove the ambiguity by an extra beam splitter.
The proposed approach inherits the simplicity of Petz’s method and avoids the laborious
and time-consuming task of moving the screen by using a mechanical system.
In order to achieve an accurate measurement, another two important problems are studied
in this thesis, namely the incident light location problem and the accurate screen pose
measurement problem. The incident light location problem is solved by the structure light
pattern method, or more precisely, the hybrid pattern method. The accurate screen pose
measurement problem, which has been little studied before, however, is extremely
important for accurate measurement. Compared with the conventional
perspective-n-points (PnP) method, a new multiple-reflections method with an auxiliary
chessboard mirror is developed to reach higher accuracy for the screen pose measurement.
A simulation of the pose measurement is conducted to prove the correctness of the multiple-reflections method.
The proposed approach uses the cross point of two 3D lines as the measured surface point.
Considering the fact that two 3D lines do not cross in most cases, an optimal guess of the
surface point is used. Furthermore, the distance of two lines is proposed as an estimation
of the measurement error. Additionally, with the calculated surface point, a quick method
to calculate the surface reflectivity is proposed with a simplified BRDF model and a
screen view angle – intensity compensation.
To prove the feasibility and accuracy of the proposed approach, a measurement system is
designed and implemented. Four calibrations of one camera and two screens are
conducted in order to reduce the measurement error, namely camera geometry calibration,
camera radiometry calibration, screen intensity calibration and screen view angle –
intensity calibration. Then three common objects with specular surfaces are measured
with the measurement system. The results show that the system reaches a relatively high
precision with 0.1mm error in average and that the measured surface reflectivity is able to
help identify potential flaws on the surface well. It can be concluded that the proposed
approach to solve the depth ambiguity in specular surface measurement is feasible and
accurate.
Post a Comment