Recognising bottles from their shape
The shape of the bottle has always been a main attribute for the recognition of bottles in the Tomra machines. In the Tomra T600 machine introduced in 1997 earlier shape sensing techniques were replaced with a CCD camera and a patented optical system for high contrast contour imaging of the bottle.
SINTEF was involved in developing the image analysis and pattern recognition software for this machine. The challenges in this development were among others:
- Develop shape features that are tolerant to the natural shape variations, e.g. due to manufacturing tolerances and buckling of thin PET bottles
- Robust statistics that tolerate lack of information in regions disturbed by loose labels, hanging patent corks, etc.
- Sub pixel measurement accuracies
- Fast processing and classification within a large set of reference shapes
- Easy and fast programming of new reference shapes
The same camera is used to monitor the insertion and transportation of the bottles through the machine. Also this task requires robust real time shape recognition since we want immediate discrimination of bottles from any other objects inserted into the machine.
If you are interested in more information please contact
Tom Kavli
.
|
We can see the bottle imaged directly and through the shape system. |
