基于机器视觉的药品标签识别及分拣系统研究

发布时间：2018-01-31 05:44

本文关键词： 机器视觉 SIFT图像匹配字符识别分拣设计运动仿真　出处：《吉林大学》2017年硕士论文　论文类型：学位论文

【摘要】：静脉滴注是一种重要的临床药物治疗手段,由此带来巨大的输液配药工作量,为了减轻医护人员的工作负担,提高输液配药质量,一类可以替代医务人员完成输液药品配制的自动配药机器人逐渐被引入各大医疗机构。针对自动配药机器人的药品标签准确识别和对错误药品的有效分拣问题,本文引入机器视觉技术,对药品标签识别与药品分拣做了相关研究。药品标签识别分为四个关键步骤:一、药瓶轮廓拾取,二、药瓶图像平面化矫正,三、国药准字编号定位与提取,四、标签字符识别。利用Canny算子检测药瓶轮廓边缘,结合竖直模板检测和图像投影法分割出药瓶图像。根据小孔成像原理,建立圆柱状药瓶成像的平面展开数学模型,有效地将相机视野范围内的药瓶图像展开成平面,同时消除药品标签字符畸变。利用SIFT图像匹配算法完成对标签图像中国药准字编号的快速定位,结合字符笔画连通域的查找与合并算法,实现国药准字编号字符串图像提取。提出基于统计特征的候选字符提取算法以及两种分别基于BP神经网络和异或增强算法的字符识别算法,引入标签字符识别一票否决制以保证药品识别结果的准确性,并通过实验论证药品标签识别算法的可行性。针对错误药品的分拣作业特点展开分拣机构设计与仿真分析。根据药瓶在传送带上的运动特点以及实际药品分拣需求,设计出用于拾取药瓶的抓手和承托抓手移动的两自由度圆柱坐标手臂机器人,并设计相应的气动系统回路。利用ADAMS对分拣机构进行整体运动仿真,证明结构设计合理,能够正确执行分拣动作。对抓手的闭合速度和夹持力等关键性能参数进行仿真分析,结果表明抓手结构设计可以满足使用需求,也为下一步的优化设计奠定基础。
[Abstract]:Intravenous drip is an important clinical drug treatment method, which brings a huge amount of infusion, in order to reduce the workload of medical staff and improve the quality of infusion. A kind of automatic dispensing robot which can replace the medical personnel to complete the dispensing of infusion drugs is gradually introduced into the major medical institutions. The problem of accurate identification of the drug label and effective sorting of the wrong drugs for the automatic dispensing robot. This paper introduces machine vision technology to study drug label recognition and drug sorting. Drug label recognition is divided into four key steps: first, bottle contour pickup, second, drug bottle image planarization correction, three. Canny operator was used to detect the contour edge of the bottle, combined with vertical template detection and image projection method to segment the bottle image. According to the principle of keyhole imaging. The plane expansion mathematical model of cylindrical bottle imaging is established, and the bottle image in camera field of vision is effectively expanded into a plane. At the same time, the distortion of drug label characters is eliminated. SIFT image matching algorithm is used to complete the fast localization of Chinese drug registration number of label image, combined with the algorithm of searching and merging the connected domain of character strokes. This paper proposes a candidate character extraction algorithm based on statistical features and two character recognition algorithms based on BP neural network and XOR enhancement algorithm respectively. In order to ensure the accuracy of drug identification results, the label character recognition one vote veto system is introduced. The feasibility of drug label identification algorithm is demonstrated through experiments. The sorting mechanism design and simulation analysis are carried out according to the characteristics of the sorting operation of wrong drugs. According to the movement characteristics of drug bottles on the conveyor belt and the actual drug sorting needs, the design and simulation of the sorting mechanism are carried out according to the characteristics of the medicine bottles on the conveyor belt. Beg. A two-degree-of-freedom cylindrical coordinate arm robot is designed for picking up the bottle and supporting the gripper, and the corresponding pneumatic system loop is designed. The whole motion simulation of the sorting mechanism is carried out by using ADAMS. It is proved that the structure design is reasonable and the sorting action can be carried out correctly. The key performance parameters such as the closing speed and holding force of the gripper are simulated and analyzed. The results show that the gripper structure design can meet the need of use. It also lays the foundation for the further optimization design.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：R955;TP391.41

【相似文献】