|
In the field of factory
automation, robot vision systems are less employed
than expected, because they often need enormous
labor to make computer programs for vision systems
to work. In general, special programs must be developed
for the applications in which they need gray level
processing, especially in the case of using a partial
silhouette of an object hidden partially due to
overlap, illuminative noise, or zoom. Thus, a general
purpose vision system is needed.
Toward such a general
purpose vision system, many model based vision approaches
have been studied to recognize and locate industrial
workpieces by matching a geometrical model given
a priori with the geometrical features extracted
from a scene image. However, the conventional approaches
using only salient geometrical features tend to
have many miss-matching cases. Therefore, they often
need special verification algorithms for individual
applications.
This paper presents
a two-dimensional model based vision system with
a general verification process using neural network
technique. The results of our experiments show the
proposed idea is useful.
@
|
