by Robert Sablatnig
Abstract:
To be acceptable in industry, vision systems must be inexpensive, within the speed of the production-line flow, and very accurate. While visual inspection is high in potential, at present the design and implementation of automatic visual inspection systems is labor-intensive. In addition, most of the visual inspection systems are developed in isolation with no systematic approach. Increasing flexibility to allow the inspection of parts whose positions are less constrained is desirable. \\ This work aims to show a systematic automated visual inspection concept that separates the detection of primitives from the model-based analysis process. This separation is obtained by defining a general analysis graph for inspection, containing detail relations that represent detection algorithms. Together with an object-specific description, defined in a socalled description language, the analysis graph is instantiated. Existing pattern recognition software is re-used in the detection stage and therefore the use of any detection algorithm is possible without changing the analysis. The concept can be seen as a ``recipe'' for solving industrial applications, stating which kind of decision have to be made at which stage. \\ An industrial application of the concept, for which industrial constraints have to be considered, is shown in the example of an automated visual inspection system for analogue watermeters used for calibration. Results concerning time, accuracy, and reliability of the specific inspection task are given. The flexibility of the concept is demonstrated by testing the analysis process with the description of other instruments (a hygrometer and a clock), which is performed by adapting the analysis graph but without changing the detection algorithms.
Reference:
A Highly Adaptable Concept for Visual Inspection (Robert Sablatnig), Technical report, PRIP, TU Wien, 1997.
Bibtex Entry:
@TechReport{TR046,
author = "Robert Sablatnig",
institution = "PRIP, TU Wien",
number = "PRIP-TR-046",
title = "A Highly {A}daptable {C}oncept for {V}isual
{I}nspection",
year = "1997",
url = "https://www.prip.tuwien.ac.at/pripfiles/trs/tr46.pdf",
abstract = "To be acceptable in industry, vision systems must be
inexpensive, within the speed of the production-line
flow, and very accurate. While visual inspection is
high in potential, at present the design and
implementation of automatic visual inspection
systems is labor-intensive. In addition, most of the
visual inspection systems are developed in isolation
with no systematic approach. Increasing flexibility
to allow the inspection of parts whose positions are
less constrained is desirable. \\ This work aims to
show a systematic automated visual inspection
concept that separates the detection of primitives
from the model-based analysis process. This
separation is obtained by defining a general
analysis graph for inspection, containing detail
relations that represent detection
algorithms. Together with an object-specific
description, defined in a socalled description
language, the analysis graph is
instantiated. Existing pattern recognition software
is re-used in the detection stage and therefore the
use of any detection algorithm is possible without
changing the analysis. The concept can be seen as a
``recipe'' for solving industrial applications,
stating which kind of decision have to be made at
which stage. \\ An industrial application of the
concept, for which industrial constraints have to be
considered, is shown in the example of an automated
visual inspection system for analogue watermeters
used for calibration. Results concerning time,
accuracy, and reliability of the specific inspection
task are given. The flexibility of the concept is
demonstrated by testing the analysis process with
the description of other instruments (a hygrometer
and a clock), which is performed by adapting the
analysis graph but without changing the detection
algorithms.",
}