【作者】 杨伟强；

【导师】 刘佐民；

【作者基本信息】 武汉理工大学 ， 机械电子工程， 2010， 硕士

【摘要】 医用输液是一种在临床应用广泛的重要的制剂之一,溶液异物的检测直接关系着患者的生命安全。因此,输液可见异物在线高效自动检测成为现代企业实现规模化生产的迫切需求。本文基于图像序列的输液可见异物检测系统以输液中所含的可见异物为研究对象,展开了输液可见异物的在线检测,对整个系统的构架与开发作了详细的设计和阐述,并重点研究了其中所涉及的视觉检测技术,包括Mean shift算法在其中的应用。首先,论述了机器视觉在工业检测过程中的应用,分析了目前存在的多种机器视觉系统,比较了各自的特点。在对机器视觉系统中通用的照明技术、图像获取技术进行了全面的论述和对各种光源的性能指标进行分析比较后,给出了多种实用的照明方式,设计了视觉传动系统、照明系统、剔除模块及电气自动控制系统；在此基础上,构建出了输液可见异物检测系统总体方案及其流程图。其次,开展了输液成像系统理论及其算法的研究,文中分析了成像系统的标定,运动检测及可见异物识别的处理过程,采用线性模型来进行相机标定。针对输液中被检测微粒的运动不确定性,结合背景模型法和帧间差分法,将归一化的互相关系数作为评价图像差异的依据,通过帧差法获得背景模型,然后将所有帧与所得背景模型相减,即得各帧中的运动目标。为了提高识别精度,本文运用均值平移跟踪器识别出下落的可见异物的改进算法。为了验证算法的实际可行性,在Visual C++和MATLAB平台上验证了该算法的有效性和可行性。最后,开发出一套输液可见异物智能检测实验软件。更多还原

【Abstract】 Medical infusion is a wide range of important clinical application. The detection of foreign objects in the solution directly related to patients’safety. Therefore the efficient automated detection of foreign objects online can be seen as the urgent need of large-scale production for a modern enterprise. The research object of the detection system based on image sequence in infusion are the visible foreign objects contained in the infusion,which can implements smart visible foreign objects inspection online. In this paper, the entire system structure was designed and described.In the structure, the technology and the software implementation are mainly researched, including the applications of the Mean shift algorithmFirstly, the application of machine vision in industry inspection is represented and the type of machine vision system is analyzed, whose characteristics are compared. A variety of practical ways in the lighting before lighting technology and image acquisition technology are a comprehensive discussion in machine vision system. The visual transmission system, lighting system, remove module and electrical automatic control system are designed. This paper shows the overall design program of foreign objects detection system in infusion.Secondly, the infusion imaging system theory and its algorithm are researched. The camera calibration, motion inspection and the identity of foreign objects are discussed, and a linear model for camera calibration is used.As for the movement of particles uncertainty, we combined with background model method and the inter-frame difference methodFor the small foreign objects motion detection, A normalized cross correlation is used to obtain the frame, based on that the background is estimated from the first frame by inter-frame difference, and small moving objects are extracted from each frame. And the normalized cross-correlation coefficient based on the difference images is as an evaluation to obtain the background model by the frame difference method. Then all the frames and the background model derived from subtract to have the frame of the moving target In order to improve identification accuracy, an adapted mean shift tracker is adopted to find falling foreign objects. In order to verify the practical feasibility of the algorithm, we have verified the effectiveness and feasibility of the algorithm. in Visual C++and MATLAB platformFinally, the software for inspection of foreign objects in transfusion bottles was developed更多还原