面向杭白菊采摘机器人的目标识别定位系统研究

发布时间：2018-01-20 16:43

本文关键词： 杭白菊摄像机标定模糊聚类立体视觉　出处：《浙江工业大学》2016年硕士论文　论文类型：学位论文

【摘要】：杭白菊作为一种传统的天然饮品,现今受到全世界越来越多的人青睐,销量也在逐年递增。为降低采收人力成本,提高作业效率,实现杭白菊自动化采摘的作业要求,研究杭白菊采摘机器人结构以及果实的检测识别及定位技术,已成为我国杭白菊生产种植全程机械化、智能化的重要环节。本课题基于机器视觉技术,针对杭白菊采摘机器人的应用需求,结合果蔬采摘机器人研究现状,运用机械系统设计理论、机器学习、数字图像处理等,提出了基于机器视觉的杭白菊识别与定位系统。主要研究内容和成果如下:(1)对杭白菊图像的预处理方法进行了研究。在HSV空间上,采用基于矢量中值滤波和形态学方法去除噪声、孔洞、毛刺等,以及对双目图像对进行图像校正处理。(2)针对杭白菊图像的特点,在传统的模糊聚类分割算法基础上,提出了基于S分量的快速模糊聚类分割方法。利用基于S分量的快速模糊聚类分割方法使传统的模糊聚类分割算法在时间效率上得到了提高,减少了迭代次数和聚类中心计算次数,处理一幅720像素×480像素图像的平均时间为0.4s,正确识别率能达到85%。(3)依据杭白菊采摘机器人对视觉系统的要求,设计了一套基于双目立体视觉的定位系统,并基于摄像机的成像几何模型,分析了双目立体视觉系统的定位原理。选用张正友平面标定法结合Matlab提供的标定工具箱对双目立体视觉系统进行了标定。实验证明,利用自制的黑白棋盘格和APP标定工具箱进行标定,可以准确、方便地获得双目立体视觉系统中两台摄像机的内外参数及畸变系数。(4)针对杭白菊果实区域特点,采用基于质心的特征匹配方法,将质心作为特征匹配的特征点,结合前期图像分割的结果,计算各个果实区域的质心,通过采用基于质心和极线约束的特征点匹配方法对同名点对进行匹配,从而实现同名果实区域的匹配,完成果实的准确定位。通过定位精度试验发现,当基线为50mm,目标离基线的距离深度为150-450mm,计算出来的距离与实际测量距离之间的误差均控制在14mm以内,能满足机器采摘的精度要求。(5)对杭白菊采摘机器人系统进行了整体设计,详细介绍了采摘机器人的基本结构,包括机械臂、采摘末端执行器和计算机控制系统等。并根据杭白菊采摘机器人的作业要求和作业流程,基于Windows开发环境,利用Microsoft Visual Studio 2013开发工具、OpenCv和MFC(Microsoft Foundation Classes)库对其识别与定位软件系统进行了设计。综上所述,本文搭建了适合杭白菊果实采摘的识别与定位系统,研究了较为合理的杭白菊果实图像预处理、杭白菊果实区域分割、识别、提取及定位方法,这些方法能为采摘机器人视觉系统的识别和定位提供一定的参考,为采摘机器人的进一步研究打下了基础。
[Abstract]:As a traditional natural drink, chrysanthemum is favored by more and more people all over the world, and its sales are increasing year by year. To achieve the requirements of automatic picking of chrysanthemum, to study the structure of picking robot and the detection, identification and location technology of fruit has become the mechanization of Chinese chrysanthemum production. This topic is based on machine vision technology, according to the application needs of Hangzhou chrysanthemum picking robot, combined with the research status of fruit and vegetable picking robot, using mechanical system design theory, machine learning. Digital image processing, a machine vision based recognition and location system of Hangzhou Chrysanthemum is proposed. The main research contents and results are as follows: 1) the pretreatment method of Hangzhou Chrysanthemum image is studied in HSV space. Based on vector median filter and morphological methods to remove noise, holes, burrs, and binocular image pairs for image correction. 2) the characteristics of the chrysanthemum image. Based on the traditional fuzzy clustering segmentation algorithm. The fast fuzzy clustering segmentation method based on S component is proposed, and the traditional fuzzy clustering segmentation algorithm is improved in time efficiency by using the fast fuzzy clustering segmentation method based on S component. The number of iterations and the number of cluster center calculations are reduced, and the average processing time of a 720 pixel 脳 480 pixel image is 0.4 s. According to the vision system requirements of Hangzhou chrysanthemum picking robot, a positioning system based on binocular stereo vision and an imaging geometric model based on camera are designed. The positioning principle of binocular stereo vision system is analyzed. The calibration of binocular stereo vision system is carried out by using Zhang Zhengyou plane calibration method combined with the calibration toolbox provided by Matlab. Using the self-made black and white checkerboard and APP calibration toolbox to calibrate, it can be accurate. The internal and external parameters and distortion coefficient of the two cameras in the binocular stereo vision system were obtained conveniently. According to the characteristics of the fruit region of Hangzhou Chrysanthemum, a centroid based feature matching method was adopted. The centroid is taken as the feature point of feature matching, and the centroid of each fruit region is calculated by combining the result of image segmentation. The matching method of feature points based on centroid and polar line constraint is adopted to match the pairs of points of the same name. In order to achieve the matching of the fruit region of the same name and complete the accurate location of the fruit, it is found that when the baseline is 50 mm, the distance depth from the target to the baseline is 150-450 mm. The error between the calculated distance and the actual measuring distance is controlled within 14mm, which can meet the precision requirement of machine picking.) the whole system of Hangzhou Chrysanthemum picking robot is designed. This paper introduces the basic structure of picking robot in detail, including robot arm, picking end actuator and computer control system, and according to the operation requirements and operation flow of Hangzhou chrysanthemum picking robot. Based on Windows development environment, using Microsoft Visual Studio 2013 development tools. OpenCv and MFC(Microsoft Foundation Classes-based software system are designed. In this paper, a suitable identification and location system for the picking of Hangzhou white chrysanthemum fruit was set up, and the more reasonable image preprocessing, segmentation, recognition, extraction and location methods of Hangzhou white chrysanthemum fruit were studied. These methods can provide a certain reference for the identification and location of picking robot vision system and lay a foundation for the further research of picking robot.
【学位授予单位】：浙江工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP242

【相似文献】