Model based reading of barcodes from low resolution CCD-images
CCD-cameras have today become a common sensing element in automation, so also in the bottle recycling machines from Tomra Systems ASA. As part of such a development Tomra wanted to investigate the possibilities and limitations in using low cost CCD-cameras for reading the bar code on bottles and beverage cans.
|
Example of a bar code decoded with the reader developed at SINTEF. The resolution in the image is around 0.3 pixels per module element. The black spot is ink from a felt pen. |
Due to the required field of view the resolution in the image would be very low, much less than the resolution in the barcode. It was thus necessary to use methods that could utilize the whole area covered by the bar code in the image and the fact that different pixels along the bars in the code generally see slightly different views of the code.
For this purpose a parametric model for the bar code, the bottle shape, and the imaging system was set up. By comparing the captured image to theoretical images computed for variations of the bar code, the most probable bar code number was established. Error rates were a few percent for images with resolution around 0.3 pixels per module in the barcode and with a selection of barcodes of relatively poor quality. The main limitation to the reading performance was dents in cans. These were difficult to estimate and include in the model.
Related article:
Model based reading of bar codes from low resolution CCD-images, Tom Kavli, NOBIM 2000
If you are interested in more information please contact
Tom Kavli
.
