【作者】 范军；

【导师】 蔡健荣；

【作者基本信息】 江苏大学 ， 食品科学， 2007， 硕士

【摘要】 随着电子技术和计算机技术的发展，智能机器人已在许多领域得到日益广泛的应用。中国是农业大国，为了提高劳动生产率、改善农民生产条件使得新概念农业机械——农业机器人的开发具有了巨大经济效益和广阔的市场前景。试验采用彩色双目立体视觉系统，从原始图像中提取出柑桔区域，对其进行处理得到其形状特征(中心、周长、面积等)。对两个摄像机分别进行标定，同时对立体视觉系统进行标定，以标定后的内外参数为基准来得到柑桔的空间位置坐标，为实现机器人采摘成熟柑桔做好前期准备工作。研究内容如下：1．对自然环境下的柑桔图像进行分析，采用Ostu自动取阈值方法对柑桔图像进行二值化处理，对传统的区域标记法进行改进，减小了图像预处理所耗时间，采用设定面积阈值的方法去除图像上非柑桔区域，防止由于开运算和闭运算对柑桔形状的破坏，同时还可以去除无法由开运算和闭运算去除的非目标小区域，最后采用四邻域判断法提取柑桔区域的边界。2．采用圆形Hough变换提取柑桔形状特征时，耗时较长，同时消耗大量的存储空间。由于柑桔的真实形状接近椭圆，采用圆形来描述的柑桔形状时会出现虚假目标，柑桔形心发生偏移。为此采用遗传算法进行椭圆拟合，对椭圆的长轴、短轴、中心横坐标、中心纵坐标、角度五个参数进行二进制编码，得到遗传基因，自定义了一个导向函数来进行基因选取，遗传操作可以得到更为准确的柑桔形状特征，耗时减少一半以上。3．分析了各种标定方法适用条件。根据实际需求，由两个摄像头分别拍摄25幅标定板图像，对图像进行分析、标定块中心坐标提取、内外对应排序得到标定数据。以摄像机针孔模型为基础，根据空间坐标系、摄像机坐标系、图像坐标系之间的几何关系建立标定方程，采用最小二乘法得到摄像头的内外参数，依据文献中提出的畸变模型对所得内外参数进行校正，为下一步的柑桔空间定位提供了准确的参数。4．对双目立体视觉测距方法进行分析，前期研究要求摄像机平行放置，影响了定位精度。本文根据空间点一定在光心与像点的空间连线上这一成像的基本原理，通过计算两条空间直线的交点来计算柑桔的空间坐标。与传统方法比较，该方法试验设备简单，对环境要求低，不要求摄像机绝对平行放置。实验结果表明本文所提出的柑桔形状特征提取与空间定位方法，在1m左右的检测距离，误差在±5.5mm以内，满足实际工作需要。更多还原

【Abstract】 With the development of electronic and computer technology, intelligent robot has been widely used in many fields. China is a big agricultural country, to increase labor productivity and improve production conditions of farmers, it is very necessary to develop agriculture robot.In this paper the color binocular stereo vision system is used,we extracte citrus region from the original image, process the image for getting the shape of citrus, calibrate two cameras ,and use the calibration results to calculate the citrus space coordinates . All works we do here is a part of the project of using robot pluck ripe citrus.The content and methods:1. Natural citrus images that we captured are analyzed first; Ostu threshold segmentation method is used to automatically process the citrus images into binary images. And a new regional marker method is used to mark regionals. The method reduce the time taken for image preprocessing,Area threshold method is used to remove the non-image citrus region, prevent the destruction of citrus shape because of opening and closing operation, meanwhile remove small area regions of non-target that can not be removed from opening and closing operation. Finally, a simple method is used to extract citrus’ border.2. Analysis of the merits and demerits of citrus’ shape feature extraction methods of using traditional circular Hough transform. Traditionally, Hough transform algorithm generally used to describe the round shape of citrus, and it is a better solution to the overlapped problem of citrus, but Hough transform algorithm has the disadvantages of massive storage requirement and high cost of computation. In this paper, genetic algorithm was used to get approximate elliptical parameters of citrus. Every ellipse can be described by 5 parameters which encoded as primitive code through binary coding, the formula of graphics matching rate is given for getting the global optimal solutions through genetic operation and some restrictions were added to fit citrus edge rapidly. The fitting citrus shape more close to the real shape, and the computational time and space is decrease to 1/2 and 1/20 compared with circular Hough transform algorithm.3. In this paper we analyze the different calibration methods applicable conditions. Most traditional methods of stereo camera calibration often require expensive calibration apparatuses and elaborate setups, which limits their practicability in many out-door applications. The paper proposes a more flexible method to overcome this drawback; we get the raw data from picture of calibration, and use Zhengyou Zhang’s Camera Calibration Technique to get the internal and external parameters. At last correct the parameter through nonlinear model. It does afford more accurate and reliable parameters to locate the citurs.4. Analysis of using binocular stereo vision measurement distance methods, traditional method needs two cameras placed absolutely parallel, it is very difficute to implement. Here, we use Three-dimensional geometric method to locate the citrus. Compared with the traditional method, reduce the system demand. Measuring about 1 m, the error is less than±5.5mm; Experimental results show that the proposed citrus identification and positioning technology is feasible.更多还原