Finding and Reading Barcodes
This webpage describes a project on developing computer vision algorithms to enable a blind or visually impaired user to find and read barcodes, such as 1D product (UPC) barcodes that uniquely identify packages (e.g. a can of beans).The goal is to guide the user to hold a camera (e.g. on a cell phone) close enough to a barcode to detect it (at a distance of several inches or more), and to get close enough to the barcode for the system to read it. Once the barcode is decoded, the corresponding product information can be looked up and read aloud to the user. This process is challenging because barcodes are hard to detect at a distance in cluttered scenes, and hard to resolve clearly enough to read, especially given the blur and other noise in images acquired by cameras on cell phones or other portable devices.
A dataset of 1D barcode images, as described in the first publication below, can be downloaded as a single zip file. The zip file contains two directories, clean/ and hard/ , and clean/ also has the train/ subdirectory. Our algorithm read all the clean images except for 020714293802d.jpg and 743877054195c.jpg . Of the hard images, it was only able to read 075403333574X.jpg and 075403333574Xb.jpg