Title:
DEEP LEARNING-BASED LOW-QUALITY TWO-DIMENSIONAL BARCODE DETECTION METHOD IN COMPLEX BACKGROUND
Document Type and Number:
WIPO Patent Application WO/2021/227289
Kind Code:
A1
Abstract:
A deep learning-based low-quality two-dimensional barcode detection method in a complex background, comprising the following steps: step 1, acquiring images and preparing same into a two-dimensional barcode dataset in a VOC2007 format; step 2, obtaining a two-dimensional barcode image by using AR glasses of a mobile terminal or industrial cameras at a fixed terminal; step 3, performing preprocessing on the received image by means of an existing algorithm based on sparse convolution kernel estimation; step 4, on the basis of a convolution kernel estimation value in step 3, performing the final restoration processing on the image by using a non-convex optimization algorithm; step 5, performing feature extraction on the restored image comprising a two-dimensional barcode by using a convolutional neutral network ; and step 6, detecting the obtained feature map by using an improved Faster-RCNN algorithm to finally obtain coordinates and angles of the two-dimensional barcode image and a two-dimensional barcode frame. The method can accurately detect and locate problems of the two-dimensional barcode such as fuzziness, abrasion, and uneven illumination.
More Like This:
Inventors:
HE WEIPING (CN)
WEI XIAOHONG (CN)
LI LIANG (CN)
WEI XIAOHONG (CN)
LI LIANG (CN)
Application Number:
PCT/CN2020/109776
Publication Date:
November 18, 2021
Filing Date:
August 18, 2020
Export Citation:
Assignee:
CYBER REALITY INNOVATION CENTRE (CN)
International Classes:
G06K7/14
Domestic Patent References:
| WO2016160912A1 | 2016-10-06 |
Foreign References:
| CN109165538A | 2019-01-08 | |||
| CN108416412A | 2018-08-17 | |||
| CN107066914A | 2017-08-18 | |||
| CN102346850A | 2012-02-08 | |||
| CN110610174A | 2019-12-24 |
Attorney, Agent or Firm:
NANJING ZHONGLV INTELLECTUAL PROPERTY AGENCY FRIM (GENERAL PARTNERSHIP) (CN)
Download PDF: