A Design And Implementation Of The QRCode Detection System Based On Convolutional Neural Network Algorithm

With the development of the Internet and the expansion of the Internet application scenarios,the frequency and the field of the QRcode using are expanding rapidly.In the current payment,social and other fields,QRcode technology has become an indispensable technology.Although the QRcode technology has been well developed and popularized in the above fields,but the QRcode detection algorithm under the general scene still has a lot of problems.In this paper,what we want to design is a QR code detection system under general scenes,so it can be used in AR application develop and robot SLAM technology.In this system,we require the system should be robust to deformation,illumination and distance.In view of this kind of application scene,the QR code detection algorithm used in the current product mainly has the following problems:1.Active detection.When the user need to detect QR code,the user needs to adjust the position of the camera to make the two-dimensional code imaging standard,clear,in order to complete the detection and identification of two-dimensional code.2.A high degree of dependence on a special identifier on the QR code.At present,the common QR code detection algorithm in the real scene is to increase the detection of the specific symbol in the specific location.3.Due to the limitation of the application scene,the detection and recognition of a single dimension code can only be performed for a single QR code.Therefore,the QR code detection algorithm above is not suitable for our QR code detection system.In order to solve the above problems,this paper designs a QRCode detection system based on convolutional neural network algorithm.The system can quickly and accurately identify the two-dimensional code in the range of the current camera.And it is applicable to general situations,at the same time,the system can estimate the QR code’s pose,so it can be use in 3D model reconstruction area and it is fit to our application scenarios.According to the actual test results,our system can be close to real-time when running on PC,and according to the test dataset of this paper,the detection accuracy has reached to 97.7%.