Precise localization of PDF417 code based on Fast Hough Transform

The PDF417 is a popular barcode symbology which is widely used in a huge variety of business processes. In this paper, we propose an original method for precise PDF417 code localization. It can successfully process projectively distorted images captured via the mobile device cameras. The core of this method is the analysis of the Fast Hough Transform image. This analysis is aimed to: (a) determine the line, corresponding to the vanish point of vertical symbol sides, using the RANSAC algorithm; (b) select the best pair of Hough-points corresponding to the horizontal symbol sides. We also propose the evaluation methodology for assessing the accuracy of precise PDF417 localization and a new dataset SE-PDF417-SYN-400, which consists of 400 synthesized PDF417 images and is publicly available. The accuracy of the proposed method on SE-PDF417-SYN-400 is equal to 0.948, and its error rate is about four times less than the one obtained by the popular ZXing detector. The average running times on iPhone 8 and iPhone 14 Pro Max mobile devices are equal to 77 and 34 ms per image correspondingly