【作者】 徐胜勇；

【导师】 李德华；

【作者基本信息】 华中科技大学 ， 控制科学与工程， 2012， 博士

【摘要】 植物器官在生长介质中的造型和分布被称之为植物形态构型。研究植物的形态构型对研究植物的生长规律具有重要的理论意义。在农业领域,对农作物形态构型的研究对于农业生产中合理选种、配种、套种和提高农业产量等有着重要的指导作用。如何通过计算机仿真定量地测定和分析植物形态构型的参数,是农业信息化中亟需解决的问题。随着计算机视觉技术的快速发展,高精度的三维信息获取技术可以满足此类需求。本文研究了一种专用于植物器官形态构型的三维信息获取系统。该扫描仪基于无接触式的被动扫描方式,仅需要使用普通的相机拍摄植物器官的多角度序列图像,再对序列图像进行计算分析得到植物器官的表面点的空间坐标信息。获取的表面点三维信息可以规范化为国际标准格式输出。为此,我们设计了系统的总体方案,构建了硬件平台并开发了系统软件。在此基础上,我们重点研究了三维信息获取系统中的相机定标。针对3DCS500型三维扫描仪的需求,本文研究并实现了一种自动化定标系统。我们基于相机和电机的出厂软件提供的函数库进行开发,实现了在用户界面中控制电机和相机的协调运作以采集目标图像,这简化了图像采集环节所需的手动操作；编程实现了控制点信息获取(包括控制点的图像坐标值和世界坐标值)和定标参数计算的自动化。该定标系统只需少量手动操作即可获取供扫描仪直接使用的定标参数。定标设备因为难以实现高精度的加工和安装,会导致控制点存在异常移位误差。本文提出了一种迭代调整的控制点定位算法,用于校正这种异常误差。先使用控制点检测方法获取原始的控制点图像坐标并用于计算一个初始定标参数。接着对控制点进行反投影并计算投影点到控制点的距离,与全局的反投影均方根误差比较,根据比较的结果对控制点的图像坐标进行调整。调整过的控制点图像坐标用于计算新的定标参数,再次进行反投影计算,以此循环直到收敛。该方法能以较小的时间代价,大幅提高控制点的定位精度。在平面模板平移的定标方法中,定标板的装配误差会导致定标板偏离理想位置,以致产生误差。对于这种装配误差,本文提出了基于基准图像的定标校正方法和基于控制点信息特征的迭代校正方法。在基于基准图像的定标校正方法中,我们从电动旋转台中心放下一个线锤,以背景屏为背景拍摄一张锤线图像。检测锤线并获取校正点信息,代入建立的定标板歪斜模型计算歪斜系数。使用歪斜系数校正控制点世界坐标存在的误差并计算定标参数。在基于控制点信息特征的迭代校正方法中,我们建立了一个歪斜校正模型去校正定标物的歪斜误差。基于靠近定标物安装点最近的控制点的歪斜误差是非常小的以致可以忽略不计的特性,提出了一个求解歪斜系数的优化算法,提高了非线性优化的速度和稳定性。两种定标校正方法获取的定标参数提升了三维重建结果的精度。为了降低定标设备的费用并简化定标操作流程,本文研究了一种基于旋转定标板的平面定标方法。使用电动旋转台来带动一个平面定标板,拍摄多张不同旋转角度的定标板图像。在定标图像中获取控制点信息(包括控制点的世界坐标和图像坐标,定标板的旋转角度)并代入我们建立的相机模型,计算定标参数。该定标方法只需较简单的操作即可获取高精度的定标参数。更多还原

【Abstract】 The modeling and distribution of plant organ in growth medium, which namely plant architecture. The study of plant architecture make an important influence in the theory of growth regular, which is playing an important role in agricultural production, reasonable selection, breeding, intercropping and improve production in agricultural area. So the determination of its parameters and analysis is crucial. It’s a significant issue in digital agriculture how to quantitatively analysis and research the relationship using computer simulation. With the rapid development of computer vision technology, high-precision three-dimensional information acquisition technology may fix those problems.In this paper, we realize a three-dimensional information acquisition system for plant architecture. The scanner based on the non-contact surveying is introduced, which can acquires 3D geometric information of the body surface through software processing with the body shape captured by visual equipments. The system is a high-tech product of abroad application. It can acquire automatically the coordinate information of the object surface points in 3D space, build 3D models of body surface, and save the acquired data as corresponding standard interface files. Several key problems of the scanner are emphasized:overall technical scheme of the system and camera calibration.We design and realize the 3D scanner which marked as 3DCS500. For this aim, we design the overall structure of the scanner, in which we establish hardware platform for operation stably and quickly, and we develop the system software including the user software and it’s functional module. All of those are realized with VC++development tools.We research on the automatic method of calibration for the scanner. We develop the control software based on the original software, by which the camera and electric motor are controlled congruously to acquire the calibration images. Control points are detected automatically and their information are used for processing calibration parameters. With a few manual operations, a set of implicit camera parameters can be obtained.The infinite precision of calibration equipment cause abnormal error to control points, which decrease the calibration accuracy. A novel control point detection algorithm is presented to correct this error. We detect control points in the calibration images and assign world coordinates for control point by its’spatial distribution. The initial camera parameters are calculated with the original information of control points using the Levenberg-Marquardt algorithm. Then the pixel distance between control point and it’s project is compared with the overall RMS error, by which we adjust the image coordinates of control point. The adjusted control points are utilized to compute a new parameters and the comparison is processed again until convergence. This method can improve calibration accuracy with a small time consuming.There are always exist deviation in assemble of planar object, which cause errors in calibration.We introduce two calibration correction methods to correct this deviation. In the first methos, we take an reference image which is the image of plumb line. With the reference image the correct points are abtained. Using the correct points we calculate the skew-correction coefficient matrix in the skew model, by which the error of information of the control point is corrected. In the second method, we produce a calibration correction method based on the information feature of control points. A skew correction model of the planar object is established to amend the skew bias. We find out and prove that the skew bias of the control point which is close to the fix point is micro enough to be ignored. Based on this property, the optimization algorithm is developed to estimate the skew factors quickly and steadily. The improved accuracy of information of control points improves calibration and the implicit camera parameters can be provided, which considerably benefits 3D reconstruction.The calibration equipments are expensive. So we develop a novel calibration system for the plant 3D scanner. With a few manual operations, a set of implicit camera parameters can be estimated directly using this system. Control points are detected with the refinement algorithm developed by us and numbered. The information of control points are devoted into the camera model and the camera parameters can be estimated. Many experiments denote the feasibility and the accuracy of the proposed method. Comparing with manual calibration methods, this system can satisfy the demand of the plant 3D scanner, which is convenient and dependable.更多还原