一种基于卷积神经网络算法的二维码检测系统的设计与实现

发布时间：2018-03-21 21:07

本文选题：二维码检测　切入点：通用场景　出处：《浙江工商大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着互联网的发展以及互联网应用场景的不断扩张,二维码的使用频次以及使用领域也在飞速的扩大,在目前的支付、社交等领域二维码技术已经成为了一个不可或缺的技术。虽然目前二维码技术已经在上述领域中得到了较好的发展与普及,但是通用场景下的二维码检测算法仍然没有非常好的效果,因此,限制了二维码在某些特定领域下的使用。本文研究是一种通用场景下的二维码检测系统,适用于增强现实应用的开发以及辅助机器人SLAM建图等,该系统需要二维码检测算法对形变、光照、距离等干扰因素有较高的抗干扰能力。针对该类应用场景,当前产品中采用的二维码检测算法主要有如下几个问题:1.主动性检测。当需要对二维码进行检测的时候,需要用户主动调整摄像头的位置角度使得二维码成像标准、清晰,才能完成二维码的检测识别。2.高度依赖二维码上的特殊标识符。目前现实场景中通用的二维码检测算法都在二维码生成的时候在特定位置增加了特定标志符辅助二维码检测。3.由于应用场景的限制,单次二维码的检测识别只能针对单个二维码进行。因此,上述二维码检测算法不适用本文课题研究的二维码检测系统应用的场景。为了解决上述问题,本文设计了一种基于卷积神经网络算法的通用场景下的二维码检测系统,该系统可以对当前摄像头所处的视野范围内的二维码进行快速准确的识别,适用于通用的场景下,同时具备姿态估计能力,可以对当前的检测到的二维码进行姿态估计,使得系统能在二维码位置进行三维模型重建等功能,该系统适用于本文上文所述的领域。根据实际测试结果,本文设计的基于二维码检测算法在PC上已经可以接近实时性,同时根据对本文的验证数据集的测试,检测精度达到了 97.7%。
[Abstract]:With the continuous expansion of the development of the Internet and Internet application scenarios, the frequency of use and the use of two-dimensional code field in the rapid expansion in the current payment, social and other fields of two-dimensional code technology has become an indispensable technology. Although the two-dimensional code technology has been in the field achieved good development and popularization however, the detection algorithm of two-dimensional code general scenario is still not very good results, therefore, limits the use of two-dimensional code in some specific fields. This paper is a two-dimensional code under general scene detection system, to enhance the development of practical applications and robot assisted SLAM mapping, the system needs to detection algorithm of two-dimensional code on deformation, light, anti interference ability of interference factors distance is high. According to the application scenarios, the products used in the two-dimensional code detection method Mainly has the following problems: 1. initiative detection. When the need for the detection time of the two-dimensional code, the user need to take the initiative to adjust the camera position angle makes the two-dimensional code imaging standard, clear, to complete the detection and identification of the.2. code is highly dependent on the special standard two-dimensional code identifier. When the detection algorithm of two-dimensional code universal reality in the scene are generated in the two-dimensional code in a specific position increases the specific detection of.3. marker assisted two-dimensional code due to the application of scene constraints, detection and identification of only single two-dimensional code for a single two-dimensional code. Therefore, the application of detection system of two-dimensional code detection algorithm of the two-dimensional code is not applicable to the subject of this thesis to the scene. To solve the above problems, this paper designs a detection system of two-dimensional code general scene algorithm of convolutional neural network based on the can of the current camera system The two-dimensional code view within the scope of the fast and accurate identification, suitable for general scene, at the same time with the attitude estimation ability, the two-dimensional code can be detected on the attitude estimation, the system could function in three-dimensional model reconstruction in the two-dimensional code, the system is applicable to herein above. According to the actual test results, this paper designs the detection algorithm of two-dimensional code is almost real-time based on the PC, at the same time according to the verification of the data in the test, the detection accuracy reached 97.7%.

【学位授予单位】：浙江工商大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP183;TP391.44;TP274

【参考文献】