基于机器视觉的马赛克分拣系统研究

发布时间：2018-06-23 13:58

本文选题：图像处理 + 马赛克　；参考：《广东工业大学》2017年硕士论文

【摘要】：马赛克作为一种常用在露面墙壁、地板外观等方面的装饰品,以其艳丽明亮的风格,在装饰领域受到越来越多人的青睐。然而现阶段马赛克分拣工艺大量依靠人工,劳动成本投入大效率低,另有一部分分拣工艺采用传统传感器组合检测方法,来识别检测分拣目标,虽然一定程度上能提高分拣系统的自动化性能,但无法满足马赛克缺陷检测要求,随着计算机图形技术的发展,采用相机来代替人眼的机器视觉技术越来越成熟,其具有可视化、无损、实时、通用性强等特点,应用在马赛克分拣领域,弥补了人工与传感器组合检测性能的不足。本文在查阅大量相关领域文献资料的基础上,基于机器视觉知识展开了对马赛克目标的分拣技术研究,过程中提出了一整套视觉识别与检测方案,并搭建实验平台,对研究对象进行了识别与缺陷检测实验。本文的主要内容包括以下几个方面:1)从马赛克识别与分拣工艺出发,设计系统方案,搭建实验硬件与控制平台,根据系统所处的实验环境,分析计算各单元零部件参数要求,选择合适型号的产品。2)提出了一整套包含图像灰度处理、图像滤波、图像形态学处理、图像分割等一系列图像处理技术的视觉识别方案,基于此视觉识别方案提取马赛克目标轮廓与表面灰度直方图特征,依据图像特征分别采用轮廓匹配与灰度值匹配来识别图像中马赛克目标与检测缺陷。3)提出了一种基于运动距离来触发相机的取图模型,根据传送带移动距离、相机视野范围、目标尺寸三者来计算相机触发节点,达到对移动传送带上的目标物体不遗漏且尽量少重复取图的目的。4)通过基于机器人的相机内外参标定,求解相机像素坐标系到机器人坐标系之间的转换矩阵,再结合传送带编码器反馈的位置数据,实现对传送带上的目标实时跟踪与抓取任务。本文采用机器视觉技术来对马赛克目标进行识别分拣,经多次反复实验证明该马赛克分拣系统具有准确率高、实时性强和速度快等优点,且具备良好的可扩展性能,能应用于其他具有明显特征的非马赛克产品的识别与分拣任务。
[Abstract]:Mosaic as a common decoration in the appearance of walls, floors and other aspects, with its brilliant and bright style, in the decoration field by more and more people's favor. However, at the present stage, the mosaic sorting process relies heavily on labor, and the labor cost is large and inefficient. Another part of the sorting process uses the traditional sensor combination detection method to identify the sorting targets. Although the automatic performance of sorting system can be improved to some extent, it can not meet the requirements of mosaic defect detection. With the development of computer graphics technology, the machine vision technology that uses camera to replace human eyes becomes more and more mature. It has the characteristics of visualization, nondestructive, real-time and versatility. It is used in the mosaic sorting field and makes up for the deficiency of the performance of the combination of artificial and sensor. On the basis of consulting a lot of literature in related fields, this paper studies the sorting technology of mosaic target based on machine vision knowledge, and puts forward a set of vision recognition and detection scheme, and builds an experimental platform. Experiments on identification and defect detection were carried out. The main contents of this paper include the following aspects: 1) from the mosaic identification and sorting technology, design the system scheme, build the experimental hardware and control platform, according to the experimental environment, analyze and calculate the parameters of each unit. (2) A whole set of visual recognition schemes including image grayscale processing, image filtering, image morphology processing, image segmentation and so on are proposed. Based on this vision recognition scheme, the feature of mosaic target contour and surface gray histogram is extracted. According to the image features, the mosaic target in the image is identified by contour matching and gray value matching. (3) A model based on moving distance to trigger the camera is proposed. According to the moving distance of the conveyor belt, the range of the camera field of vision is obtained. The target size is used to calculate the camera trigger node, so that the target object on the moving conveyor belt is not omitted and the image is retrieved as little as possible.) the camera is calibrated based on the robot's internal and external parameters. The transformation matrix between camera pixel coordinate system and robot coordinate system is solved, and the real-time tracking and grasping task of the target on the conveyor belt is realized by combining the position data feedback from the conveyor belt encoder. In this paper, the machine vision technology is used to identify and sort the mosaic target. It is proved by repeated experiments that the mosaic sorting system has the advantages of high accuracy, high real-time and high speed, and has good expansibility. It can be used in the identification and sorting of other non-mosaic products with obvious characteristics.
【学位授予单位】：广东工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】