The changing of international market demand, hardware
facilities and communication technologies lead to
manufacturer step on actual time information,
human-robot collaboration, real time feedback and
Co-design with different studio. However, core
technologies are possessed by foreign companies, causing
difficulties of implementing smart manufacturing. In
recent years, Taiwan’s
government-industry-university-institute alliance
actively develop basic technologies to improve domestic
company’s critical technologies.
One
of basic technologies in smart manufacturing is machine
vision. In this field, template matching technology can
be applied to manufacturing, medical industry,
entertainment industry, and so on. However, industrial
level template matching technology requires high
performance in recognition speed and accuracy, causing
monopolization of foreign companies. This research will
focus on the development of template matching, a core
technology of machine vision. The recognition speed of
the template matching software developed in this
research is comparable with commercial software.
In this research, light field distribution is the
important reason of environment factor. To deal with
light field distribution, Normalized Cross Correlation
(NCC) will be the basic of calculation algorithm.
However, the massive calculation of this method will
lead to low recognition speed, which is not acceptable
to industrial user. As a result, previous research has
developed various simplified models that successfully
increase computational speed. However, the decision of
key parameters in simplified models hasn’t been
discussed in previous research. This search proposed a
simplified template recognition model based on Fast
Normalized Cross Correlation (FNCC). To avoid the
uncertainty of parameters and the ignorance of feature
of FNCC, the simplified model of this research are
constructed by basis function and add method. Parallel
Operation of Signal Instruction Multiple Data (SIMD) is
implemented to accelerate recognition. Results show that
the template matching software proposed in this research
have comparable recognition speed in comparison with
commercial software.
Keywords: Template Matching, Normalized
Cross Correlation, Basis Function, Single Instruction
Multiple Data
