Dynamic QR Code Recognition System Based On Convolutional Neural Network

The QR code identification system is an important part of the logistics sorting system,and its recognition efficiency affects the sorting efficiency of the logistics sorting system to a certain extent.At present,the QR code recognition system of most domestic logistics sorting centers is manually identified by manual QR code recognizer,and gradually developed into automatic dynamic recognition using high-speed cameras.While the recognition efficiency has been improved,the investment of equipment has been greatly increased,which has hindered the popularization of its system.In this paper,the convolutional neural network and motion blur theory are studied.Combining the target detection technology and digital image processing technology in machine vision,a dynamic QR code recognition system based on convolutional neural network is developed.The system is low cost and high efficiency.The real-time recognition function of the dynamic QR code is realized,which has certain application value.The work of this paper mainly includes the following three parts:(1)Development of dynamic QR code location algorithm.The biggest difficulty of this system is how to recover fast and high quality motion blurred images.In view of the timeconsuming characteristics of the deblurring algorithm,in order to meet the real-time requirements of the system operation,this paper adopts the idea of first positioning and segmenting the QR code region and then restoring it.At present,deep convolutional networks are widely used in the field of target detection,which has higher accuracy and real-time performance than traditional target detection methods.Based on the current excellent target detection algorithm and its advantages and disadvantages,this paper proposes a dynamic QR code localization algorithm based on convolutional neural network.Aiming at the usage scenarios of this system,a method for establishing dynamic QR code image dataset is proposed,and the initial parameters in the training process are improved.Finally,the accurate positioning of the dynamic QR code was successfully achieved.(2)Research and implementation of dynamic QR code recovery algorithm.Firstly,the fuzzy fuzzy distance based on differential autocorrelation and the fuzzy angle estimation algorithm based on quadratic transformation are implemented,and the fuzzy kernel of motion blurred image is estimated.Secondly,the commonly used efficient deblurring algorithm is simulated and tested,and the deblurring algorithm based on Wiener filtering is selected according to the actual performance.Then,in order to further improve the quality of the restored image,a loop filtering algorithm and an Otsu binarization algorithm are introduced.Finally,the recovery of the dynamic QR code image is achieved.(3)Implementation of a dynamic QR code system.Software part: Under the Linux system,based on the deep learning framework TensorFlow,the algorithm for locating the dynamic QR code in the image is realized.In the Windows+QT development environment,combined with the image processing tool library OpenCV,the motion blur recovery algorithm for dynamic QR code images is realized.Hardware part: comprehensive performance,price completed the selection of industrial cameras,optical lenses and servers,and designed the structure of the auxiliary light source,completed the hardware platform of the system.Finally,the dynamic QR code recognition system was tested.The experimental results show that the system runs at a certain speed and has good recognition accuracy under the condition of ensuring real-time performance.