Intelligent pattern recognition in 3D vision for robot control
Toshio Fukuda · 2003
Proposes a recognition method by which 3D information of overlapped objects can be constructed, taking into account the fuzzy quantities of linearity and roundedness of image data. The recognition approach was based on grey-level image processing as well as color and binary image processing. Grey-level image data were used for classifying the image patterns into the specific candidate group. Then the other data are obtained by changing the attitude of the objects slightly, so that the 3D image data can be reconstructed for pattern recognition and understanding. An expert system for the pattern recognition and the understanding of objects is constructed based on a production system. The system gives commands to rotate the objects in a certain direction to increase the image information about the objects until they can be identified.>