Research On High-performance Multi-barcode Detection Methods In Complex Scenarios

Barcode technology is playing an increasingly important role in daily production and life due to the rapid development of industry,commerce,and logistics.With the popularization of smart terminal equipment,barcode technology and related algorithm research have been developed.Barcode scanning can be decomposed into two parts: detection and decoding.Accurate detection can greatly improve the speed and accuracy of decoding.At present,the existing barcode detection algorithms still need to be improved when detecting difficult barcodes in low-quality images.This dissertation proposes different barcode detection algorithms with strong scene adaptability based on digital image processing and computer vision.For a one-dimensional barcode with a relatively fixed structure,barcode intensity operators based on the second-order structure matrix and the gradient direction information entropy are respectively defined through the analysis of gradient characteristics,based on which a dual-calculator barcode detection algorithm is proposed by utilizing parallel calculation.Post-processing steps such as filtering and candidate region selection are designed to improve detection effect,while rotation is also applied to improve the accuracy of decoding.The dual-calculator barcode detection algorithm can triumph difficult barcode detection tasks such as uneven illumination,blur,distortion,low contrast,and large field of view conditions while owning batch processing ability.Different from one-dimensional barcodes,the structural characteristics of twodimensional barcodes vary in a wide range.Aiming at this problem,another algorithm based on deep learning is proposed to handle general barcode detection task.The SSD network is used as framework while the backbone network is replaced with PeleeNet to reduce model computation.Meanwhile,a front-end filtering module and a deconvolutionresidual module is designed to enhance feature extraction ability.The loss function and training parameters are also optimized.The algorithm is tested on a dataset made by an industrial camera,exhibiting an extremely high accuracy.The proposed methods have been tested on publicly available one-dimensional barcode data sets such as Muenster and ArteLab,and the Jaccard index as well as detection rate are higher than the existing barcode detection algorithms,while meeting the real-time requirements.