【作者】 张东波；

【导师】 张宾；

【作者基本信息】 中国农业大学 ， 机械制造及其自动化， 2014， 博士

【摘要】 传感技术是实现精细农业的基础,在自动化农业生产作业中需要对作业对象目标的空间信息进行检测进而用于对作业执行机构的控制,机器视觉作为潜在的、低故障、低风险、非接触式测量技术非常适宜于农业环境的应用,大多数的农业生产作业需要执行机构与作业对象间的相对运动,本文以机器视觉技术结合农业生产作业中的执行机构与目标物的相对运动关系提出了一种基于视点移动的植物三维信息检测和一种直接基于图像特征的摄像机相对于目标物的视觉伺服定位控制策略,主要研究内容和结果如下：(1)对基于色彩空间的绿色植物信息分割提取方法进行了研究。以手动提取绿色植物的方法制备植物信息分割提取的样本图像集,分别以植物欠分割率、植物过分割率、植物错分割率和图像处理时间为分割性能的比较参数对植物信息分割算法进行了试验对比研究。设计了一种结合HSV颜色空间下的固定阈值分割和超绿减超红(3G-2.4R-B)图像数据Otsu自动取阈值分割的植物图像分割策略。(2)研究了图像对应点的搜索匹配算法。采用绝对误差和作为区域匹配的误差度量,对图像中不同位置点的对应点搜索在作用图像数据不同、匹配模板类型不同和匹配模板大小不同的情况下的匹配效果进行了试验研究,设计了依据匹配点的位置的不同而不同的差异性的匹配算法。(3)研究了摄像头在X-Y平面内运动时的植物三维信息检测算法。以作物为试验对象对摄像头在X-Y平面内不同运动参数和位置参数对植物三维信息检测的结果的影响进行了试验研究,试验结果显示,摄像头的移动距离越大,摄像头与目标物的距离越近,检测结果越好。(4)研究了摄像头沿光轴方向移动时的目标与相机的深度信息检测方法。对摄像头沿光轴方向移动时的目标物在摄像机投影方向的等效平面与摄像头的距离的计算方法进行了研究,对摄像头与植物的距离不同以及摄像头的移动距离不同情况下对测量结果的影响进行了试验研究,试验结果显示,摄像头的移动距离越大,摄像头与目标物的距离越近,测量的结果越好。(5)采用两相步进电机驱动的电动平移台和直流电机驱动的电动缸搭建了一个直角坐标机器人系统,基于摄像头沿光轴方向移动的目标与摄像头深度信息检测方法设计了视觉伺服控制策略。分别以三个图像特征实现对执行机构在空间内的三个方向移动定位的视觉伺服控制,以绿色作物为试验对象实现了视觉伺服定位控制功能。本文的研究结果为机器视觉在农业生产中的检测和对执行机构的控制方面的应用奠定了基础,研究对基于手眼方式的自动化施药、采摘、移栽等农业生产作业具有重要价值。更多还原

【Abstract】 Sensing technology is the basis for precision agriculture. Detect spatial information of the target, which can be used to control the the actuator, is needed in automated agricultural production operations. Machine vision is considered to be a potential, low failure and risk and contactless measuring technology which makes it suitable for being applied to agricultural circumstances. Most of the agricultural production tasks accompany the relative motion between the objects and the actuator A method of the plants’three-dimensional information detecting based on the translation of the camera and a visual servo positioning control strategy directly based on the image features was proposed, which was combined used the motion relationship and the machine vision technology. The main research contents and conclusions were as follow:(1) Plant image green information segmentation method was studied based on different color space. A set of sample images’ green information was extracted manully. Plant defective segmentation percentage, plant exceeded segmentation percentage, plant segmentat error percentage, as well as the image processing time was used as the performance parameters of different segmentation algorithm. A green information segmentation method which combined the fixed threshold segmentation under HSV color space and threshold segmentation by Otsu for the image data after the operation of Exceed Green minus Exceed Red (3G-2.4R-B) was designed.(2) The corresponding points matching algorithm was studied for images acquired from different views. The sum of absolute defferences (SAD) was choosed as the mearement error for the area maching. The maching algorithm was studied based on the experimentions which were operated on different image data, with different template size and different template types. A difference matching algorithms strategy was designed depending on the position of the matching points.(3) Plant three dimensional information detection algorithms were studied when the camera was translated in the X-Yplane. The experimentions was done with the crops as the subjects in order to find out the influence on the detected results of the plant three dimensional information while the camera was under different motion parameters and different position parameters. According to the experiment, the direction of movement of the camera has little effect to the measurment results, and there would be get better results when the moving distance of the camera becoming larger and the distance was closer between the camera and the object.(4) The depth information between target and the camera was studied while the camera was moving along the optical axis. The distance between the equivalent plane of the target in the direction of projection of the camera and the camera center would be calculated when the camera was moving along the optical axis. A standard target and a potted plant was chose as the subjects in order to find out the measurement performance under the different distance between target and the camera and the different distance of the camera moved. As a result, the larger the moving distance of the camera, the closer between the camera and the target object, the better of results of the measurement. (5) A visual servo positioning control strategy based on the method depth information measurement between target and the camera when the camera was moving along the optical axis. Three image features was directly used to control the moving and positioning of the actuator in the three directions of space. A Cartesian robot system was set up with two electric translation platforms driver by two-phase stepper motor and an electric cylinder driver by DC motor. The experiments of target’s three dimensional measurements and visual servoing control of the camera relative to the plant was done based on the robot system. A satisfying result was abtained through the experiments.Research laid the foundation for machine vision inspection and the actuator’s visual servoing in agricultural production. Meanwhile this tudy has important value to automate spraying, harvesting, transplanting and other agricultural production operations which based on hand-eye approach.更多还原