【作者】 张伟华；

【导师】 陈军；

【作者基本信息】 西北农林科技大学 ， 农业电气化与自动化， 2009， 硕士

【摘要】 随着计算机图像处理技术逐渐成熟,视觉导航将成为农用轮式移动机器人自主导航的重要发展方向。本文研究的是农用轮式移动机器人视觉导航系统的一个关键部分—导航路径的识别以及车辆导航参数的提取,是实现车辆自主导航的重要模块,对视觉导航技术的发展和推广有重大意义。由于农用轮式移动机器人在实时作业时,导航路径一般近似直线,因此,本文针对结构化道路,将道路的中间线作为车辆的导航路径进行识别并提取车体的导航参数,主要研究内容和结论如下:(1)对CCD摄像机采集的彩色图像进行预处理,并重点研究了图像阈值分割方法。实验证明迭代分割和Ostu分割均能产生较好的二值化图像。(2)在国内外学者的研究基础上,深入研究图像边缘检测。分析了基于模板卷积运算的Robets、Prewitt、Sobel、Laplacian等多种边缘检测算法,实验结果表明运用Laplacian算法能在满足实时性要求下清楚地检测到道路边缘。(3)研究一种快速有效的直线信息检测方法识别农用轮式移动机器人的导航路径。分析了地面直线路径在图像空间中的投影特点,通过改进Hough变换直线检测算法,在获取农用轮式移动机器人跟踪路径的同时直接提取了车辆的导航参数—横向偏差和航向偏差;实验结果显示横向偏差和航向偏差的平均值误差分别为2.6(cm)、-0.47(deg)。(4)采用针孔成像模型完成CCD摄像机的标定。建立了机器人坐标系和计算机图像坐标系中各对应坐标点的关系方程组,通过求解得出摄像机各内外参数。(5)以福田欧豹4040型拖拉机作为农用轮式移动机器人实验原型,采用CCD摄像机采集车辆行驶的道路图像(彩色位图),利用VC++6.0编写图像处理源程序。实验表明拖拉机在低速行驶时,本文开发的图像处理源程序能够满足实时性要求,处理速度为2.1Hz,平均每秒钟处理2帧640×480(像素)彩色位图。更多还原

【Abstract】 Gradually, with the maturity of computer image processing technology, vision navigation is going to be an important developing direction of the agricultural mobile wheeled robot. Research of the paper is to recognize the navigation route and to obtain vehicle navigation parameters which is a key section of the agricultural mobile wheeled robot navigation system and also is the important module realizing vehicle independent navigation. The study will make great sense to the development and popularity of vision navigation technology. Generally speaking, when the agricultural mobile robot is on real-time busywork, the navigation route is approximate line, therefore, based on structured road, the study about navigation route recognition of agricultural wheeled mobile robot is carried on. The major research contents and conclusions are as follows:(1) The color images collected by CCD camera are preprocessed and the methods of image threshold partition is paid more attention to. Both iteration segmentation and Ostu partition are considered to be able to provide better binary images.(2) On the base of scholars’research at home and abroad, image edge detection is deeply studied. Grounding on template convolution operation, some edge detection algorithms such as Robets, Prewitt, Sobel and Laplacian are analyzed. Experimental results prove that Laplacian algorithm can satisfy the real-time requirement and detect the rode edge distinctly.(3) A quick and effective line-detection method to recognize the navigation route of agricultural wheeled mobile robot is studied. Projection characteristics of a line on the ground in image space is analyzed and the navigation parameters—lateral and heading displacement, are gotten directly at the same time as the traced route of the agricultural mobile robot is detected by improving the algorithm of Hough transformation which is used to recognize the line. The experimental result shows that the error of lateral and heading displacement respectively are 2.6 centimeters and negative 0.47 degrees.(4) CCD camera calibration is completed with pinhole imaging model. A series of relationship equation group of corresponding points in robot’s coordinate and digit image coordinate is established, after calculation, each inner and outer parameters of the camera have been obtained.(5) The Fu Tian O Bao 4040 tractor is regarded as the prototype of agricultural wheeled mobile robot. The CCD camera is used to collect the tractor steering rode images(color bitmap) and the VC++6.0 is utilized to write image processing programs. The experiments reveals that when the tractor speed is slow, real-time requirement can be achieved through the software development procedure of this thesis, with the processing speed being 2.1Hz and two frames 640×480(pixel) color bitmaps processed per second.更多还原