【作者】 张果胜；

【导师】 沙莎；

【作者基本信息】 中南大学 ， 计算机科学与技术， 2010， 硕士

【摘要】 随着工业、军事等领域对异常检测需求的日益增大,加之高性能、高可靠性图像处理算法的日益成熟,促使机器视觉技术广泛应用于异常检测当中。针对实际需求和一些亟待解决的问题,本文开展研究,设计并实现了基于模板的异常检测系统。针对单一的异常检测算法无法普遍适用于各种应用环境和不同特征检测对象的问题,提出采用工厂策略模式对系统的异常检测算法模块进行设计,并实现了利用灰度信息配准匹配的异常检测和利用轮廓信息对准匹配两种适用于不同应用场合的异常检测方法。通过工厂策略模式和相应的配置文件,在应用环境或者检测对象的特征发生变化时,客户可以动态创建、配置和加载符合当前异常检测需求的算法,确保在当前应用中,异常检测模块能在快速、稳定、准确之间保持很好的平衡。针对利用灰度信息模板匹配的异常检测方法容易受到摄像机抖动和环境光照变化影响的问题,设计了一种具备容错能力的分块投影匹配法,实现了摄像机抖动的图像纠正；接着在介绍了利用高斯滤波器和多项式拟合消除光照变化方法的基础上,提出了实时性能更好的Wallis变换法来消除光照变化的影响；最后基于加权多区域差分矩阵进行异常决策。实践证明,在摄像机抖动量不大,光照连续变化的情况下,该方法能实现快速、准确地检测异常。针对利用轮廓信息对准匹配的异常检测方法对轮廓点分割的准确度要求很高的特点,首先对比了常用的边缘检测算法,提出采用Canny算子来准确提取图像的边缘,克服了传统边缘检测算子易受噪声干扰和粗轮廓的问题。接着基于边界追踪构建目标轮廓,同时计算轮廓的链码熵、链码空间分布熵和轮廓矩。最后先基于全局特征相似度的粗匹配在待检测图像中找到包含待检测对象的所有区域,再基于轮廓对准的细匹配进行异常判断,并精确地定位轮廓上异常的具体位置。针对注塑模具的保护需求,实现了基于模板的异常检测系统。实践证明,本文提出的异常检测算法能够保证检测的实时性和准确性,阶段性地达到了研究的目的。更多还原

【Abstract】 With the demand for abnormity detection increased day by day in industrial and military fields,high-performance and high reliability image processing algorithms become more mature, which promotes machine vision technology widely used in abnormity detection.Aimed at the actual demand and a number of issues needed to be addressed, this paper carried out the research, designed and implemented a template-based abnormity detection system.Aimed at the problem that a single abnormity detection algorithm can’t be universally suitable for various application environments and detection objects of different characteristics, we proposed the plant strategy model to design the abnormity detection module of the system and realized two kinds of abnormity detection methods:registration matching using gray information and alignment matching using contour information, which are suitable for different application situation.Through the plant strategy model and some corresponding configuration files, whether application environment or the characteristic of the detection object changes,the client can dynamically create,configure,and load the algorithm meeting the current demand of abnormity detection to ensure that the abnormity detection module can get a good balance among rapidity, stability and accuracy in the current application.Aimed at the problem that the abnormity detection method of template matching using gray information is easy to be influenced by camera vibration and illumination change, this paper firstly designed a fault-tolerant block projection matching method to realize the image correction of camera vibration, and then proposed the Wallis transformation method to eliminate illumination change, which is more real-time than the mothed using Gaussian filter and polynomial fitting introduced before. Finally, carrying out abnormity decision based on weighted multi-areas difference matrix.The practice shows that in the situation of small camera vibration and successive illumination change, this method can quickly and accurately detect abnormity.Aimed at the characteristic that high accuracy contour point segmentation is necessary for the abnormity detection method of alignment matching using contour information, this paper firstly compared common edge detection algorithm, and proposed the edge extraction method using canny operator to overcome the problem that traditional edge detection operator is easy to be influenced by noise and the problem that the contour is too thick. And then boundary tracing is used for constructing target contour and calculating chain code entropy, chain code spatial distribution entropy and contour moment. Finally, coarse matching based on global feature similarity is used for finding all areas in the image that contain the detection object, and then fine matching based on contour alignment is used for carrying out abnormity decision to precisely locate abnormity in the contour.Aimed at the protection demand of injection mold, this paper realized the template-based abnormity detection system. The practice shows that abnormity detection algorithms proposed by this paper can guarantee real-time and accuracy abnormity detection and achieve the purpose of the study by phases.更多还原