Research On Whole Scene Bar Code Image Localization And Recognition

With the popularization of e-commerce and the continuous development of the express market,a large number of barcodes are used in modern logistics to record various information about items.Obtaining barcode information in different links of logistics is an important technology to improve logistics efficiency and management.Constrained by factors such as the volume of logistics products in the actual scene and the location,it is necessary for two cameras to overlap the field of view to cover the complete item.Therefore,this article uses a binocular camera for barcode recognition.This article combines deep learning,digital image processing and other technologies to establish a full-scene barcode recognition system to achieve rapid positioning and recognition of multiple barcodes in the image.The main content of the article research is expanded from the following three parts:(1)In order to achieve accurate and rapid positioning of multiple barcodes in the image,this paper proposes a barcode positioning algorithm based on deep learning.The article is based on the Darknet53 network of YOLOv3(You Only Look Once,YOLO)and the improved loss function to retrain the barcode data set.In order to cope with the changes in the demand for bar types in the future,the data set includes various common barcodes in the current logistics industry..The target detection algorithm based on deep learning can automatically extract image features without manual interference,and optimize feature extraction,target classification,and target positioning in the same network.It has simple steps to use and high robustness.advantage.The experimental results show that the improved loss function has a faster convergence speed and a better model effect for large angle errors.Compared with the traditional barcode positioning method based on digital image processing,the article algorithm can reach the speed of barcode positioning on the WWU Muenster Barcode dataset.31 fps,the accuracy rate increased by 15%.(2)In order to realize barcode deflection angle detection,this paper proposes a barcode deflection angle detection algorithm based on deep learning.The article uses the basic network of Darknet53,adds a barcode deflection angle detection structure to its detection head,integrates barcode image positioning,deflection angle detection and classification into one network for optimization,and designs losses for positioning,classification and angle regression information respectively Function,optimized the network training structure.The experimental results show that on the WWU Muenster Barcode data set,the algorithm in this paper improves the accuracy by 5% compared with the traditional algorithm,and the real-time detection rate can reach 27 fps.In order to quickly and efficiently calibrate the bar code angle information in the data set,this paper also designs a bar code deflection angle detection algorithm based on LSD(Line Segment Detector,LSD).Based on the LSD detection result,the algorithm calculates the total length of all straight lines at the same angle,and uses the angle corresponding to the maximum value as the deflection angle of the barcode.Experimental results show that the average speed of the algorithm to detect a bar code deflection angle is 18 ms.(3)In order to realize the barcode detection in the overlapping area of the binocular camera's field of view and provide a full-scene logistics monitoring picture,this paper proposes a fast image stitching algorithm.In the YUV color space,Y represents brightness information,and UV represents color information.Therefore,Laplacian image fusion can only be performed on the Y channel,which reduces the computational complexity while ensuring the quality of image stitching.The experimental results show that compared with the traditional algorithm,the algorithm in this paper has increased the fusion speed by about 3 times,and the stitching speed can reach 22 fps while ensuring a higher stitching quality.The research method of this article is applied to barcode detection,and the rapid positioning and recognition of barcode is realized.In the barcode recognition scene based on binocular cameras,it can quickly perform image stitching,handle the abnormal situation of barcode recognition in the overlapping area of the field of view,and provide real-time logistics monitoring images of the whole scene.