MULTIDIMENSIONAL PATTERN-CLASSIFICATION OF BOTTLES USING DIFFUSE AND SPECULAR ILLUMINATION

A robust and fast method for segmenting and recognizing bottles based on features extracted only from their circular tops is presented. The approach exploits a dual (stereo) camera system that views trays of bottles illuminated by one of two spatially separated sequenced light sources. The first of these light sources produces specularly illuminated bottle caps while the other diffusely illuminates them. The images obtained using light sources that produce specular reflections are employed primarily for segmentation purposes, although two features based on mean specular intensity and bottle height are extracted from these images. The images obtained while the diffuse light sources are active are used to extract three other features which are the mean diffuse intensity, intensity variance, and intensity distribution of the bottle caps. It is shown that this dual lighting approach increases the overall reliability of the integrated system which must be capable of accurately segmenting and recognizing many bottle types. Results of test runs involving sets of hundreds of training samples and unknowns are presented and demonstrate that recognition rates in excess of 99% may be expected.