Research On Container Box Number Identification Method Based On Computer Vision

The economic and social development has promoted the rapid development of thelogistics industry.Containerized transportation is one of the main mode transportation in the logistics industry.In order to effectively control the container,the container number identification system has been widely used.It mainly realizes the identification of the box number.The box number identification system can effectively reduce the recognition time,which saves labor,improves the accuracy of the box number identification,and shortens the residence time of a single container so as to improve container transportation efficiency.This thesis mainly studies the positioning,segmentation and identification of the container box number identification method,and realizes the box number identification system based on OpenCV platform.The main work is as follows:(1)Aiming at the problem of stains and rust effecting positioning in the container picture,the image is firstly equalized to improve the quality of the picture,and then Gaussian filtering is used to denoise.Finally,the Bayesian network is used to locate the box number area.The positioned box number may need to be corrected for the tilt of the character.In this thesis,the least square method and the Radon transform are used to correct the tilted area.After the tilt correction is completed,the character segmentation is easier to implement.(2)Aiming at the problem of characteristics adhesion and fracture of the container picture,In this thesis,the method of combining the centroid method and the projection method is used to realize the segmentation of the box number.The improved centroidmethod first finds the centroid of each character,and then determines the middle position of the adjacent character centroid as the segmentation point.Finally,the projectionmethod is used to realize the segmentation of characters.Accurate segmentation can lay a foundation for character recognition.(3)Aiming at the problem of similar characters misunderstanding of box number identification,this thesis improves the traditional template matching method and combines the character structure method to realize the identification of the box number.This method first classifies the characters thatmake up the box number.All characters with similar shapes are classified into one class,and each part of the character is given different weights to ensure that similar charactersare not misunderstood.When the templatematching method is improved to identify errors,the character structure method performs secondary recognition to ensure the accuracy of recognition.(4)The thesis designs a container box number recognition system based on OpenCV platform and implemented in C++.After preprocessing the image,it realizes the location of the box number through Bayesian network.After the positioning is realized,the box number identification system completes the box number segmentation by the improved centroid method and projection method.Finally,the improved template matching method and the character structure are combined to realize the identification of the box number.The box number identification system displays the time and identification result of the box number identification in the form of interface.By testing a large number of container pictures online,the result shows that the correct rate of box number identification is improved,which proves the feasibility of the system.