Automatic Barcode Extraction for Efficient Large-Scale Inventory Management

Abstract

Barcodes are extensively used in inventory management to identify and track products and components. While the current prevalent method of reading barcodes with handheld barcode readers is suitable for retail checkout lanes and small-scale inventory management, it presents significant work efficiency and worker safety challenges for large-scale inventory management where products are usually stored on multistory racks in large warehouses or storage facilities. Such facilities present significant opportunities to improve logistics by automating the inventory management process. To ease inventory tracking while mitigating the identified issues, this research takes advantage of computer vision based barcode readers to replace traditional barcode readers and proposes three techniques to facilitate barcode extraction from video scan data. First, Harris corner detector and Hough transformation are proposed to work in unison to estimate the direction of a region including a single barcode and rotate it to an ideal state for the benefit of existing decoding algorithms. Then, an algorithm based on barcode region connectivity and geometry property is proposed to find multiple barcodes in a single image to avoid brute force searching of valid barcodes. In addition, a histogram difference based key frame selection method is also proposed to eliminate redundant information between sequential frames which helps to improve efficiency by using fewer frames. Finally, these techniques are applied to video data collected in a large logistics warehouse and the overall performance is evaluated. The experimental results show that our algorithm can achieve both satisfactory recognition rates and efficiency, and thus offers significant promise for broad deployment in automated large-scale inventory management using terrestrial and aerial robotic platforms.