【作者】 丁亚兰；

【导师】 耿楠；

【作者基本信息】 西北农林科技大学 ， 计算机应用技术， 2009， 硕士

【摘要】 本文以猕猴桃为研究对象,对猕猴桃采摘过程中目标识别与定位的关键技术进行研究,基于机器视觉技术实现了果实的有效识别和空间的初步定位,从而为猕猴桃果实的采摘提供了必要信息。取得的主要的研究成果和结论如下:(1)在基于机器视觉技术的猕猴桃果实识别中,为了确定有利于图像分割的颜色空间和颜色因子,对提取的40幅猕猴桃图像在RGB颜色空间中的颜色分量值进行了统计分析,通过对猕猴桃果实、树叶、树枝的感兴趣区域颜色特征的分析,研究表明R-B颜色因子对于猕猴桃果实分割最为有利。(2)分别采用固定阈值法和自动阈值法对R-B分量图像进行了分割,分割效率分别大于86%和81%。其中固定阈值法具有较好的分割效率和实时性;自动阈值具有较好的分割效果和一定的自适应性,但耗时较多。针对分割处理后的二值图像存在残留物问题,利用图像填充结合形态学处理的方法进行了残留物去除,取得了较好的效果。(3)根据果实采摘机器人对视觉系统的要求,确定了果实目标的边界、面积、圆形度、形心、外接矩形等作为图像识别特征参数。(4)研究了基于猕猴桃形心特征的匹配方法,利用二维图像上猕猴桃果实的形心特征,实现了双目果实图像的立体匹配。(5)在立体匹配的基础上,利用立体视觉原理实现了空间点三维信息的计算,初步确定了猕猴桃果实的空间位置。通过对猕猴桃特征点空间位置信息进行测量,实验表明:该定位方法在实际深度为800mm左右时,空间位置的计算误差为9.03mm。更多还原

【Abstract】 With Kiwi fruit as the subject, the key technology of fruit recognition and location in developing the kiwifruit picking robot is studied; the machine vision technique is used to collect necessary information for effective recognition and location of kiwifruit. The main research results are as follows:(1)In order to identify the best color spatial and color mode, based on robot vision technique in Kiwifruit recognition, the color components distilled in RGB color space of 40 pieces of Kiwifruit images are chosen for statistical analysis. With the analyses of color character in kiwifruit, leaves, branches and other important areas, the R–B color mode is the best to segment Kiwifruit;(2)The fixed and automatic thresholds are adopted respectively to segment the R-B Component images and the segmentation efficiencies are higher than 86% and 81%. After comparison, the fixed threshold method has the better segmentation efficiency and real-time performance, automatic threshold has a better segment effect and adaptability, but it consumes more time. To solve the residual problem usually existing in segmented two value images, flood fill and morphology operation method are applied to eradicate the residual preferably;(3)To meet the vision requirements of the fruit robot, outline, area, center of gravity and enclosing rectangle are confirmed as parameters for extracting the object’s characters for image recognition;(4)Develop the matching method of the Kiwifruit centroid; use the centroid characters of the fruit in 2D image to match the binocular images of Kiwifruit;(5)Using the stereo vision principle after stereo matching, calculate the 3D information of spatial points and locate the position of the fruit. With precise dimensional measurements on the position data, the experiments show that calculation error of the special position is 9.03mm when the orientation depth is near 800 mm.更多还原