【作者】 汤磊；

【导师】 王强；

【作者基本信息】 西华大学 ， 机械电子工程， 2008， 硕士

【摘要】 足球机器人是目前机器人研究领域的一个热点课题,它涉及了诸多领域的前沿研究,是一个极富挑战性的高技术密集型研究平台。视觉系统作为足球机器人四大子系统之一,是足球机器人的重要组成部分,其主要任务是通过视频设备采集场地图像,并对获取的图像进行适当处理,最终获得各目标体在场地中的位置和角度信息,从而为决策系统提供初始状态信息。衡量视觉系统的性能优劣主要体现在对目标识别的精度和速度。因此,本文的研究重点就在于如何提高系统的实时性和辨识精度。本课题在硬件方面采用三星公司的SCC-331摄像头获取图像,并通过CA-CPE-3000型图像采集卡将模拟图像转换为数字图像,该视觉系统具有采集速率较高、图像画面优越等特点,不仅便于后期的图像处理,而且可以提高系统的实时性。本文在分析了足球机器人视觉系统目前常用的几种颜色模型后,根据选择的硬件的特点,选用最为简单的RGB颜色模型进行图像处理,大大减小了系统的计算量。在分析了几种常用色标的基础上,提出了队标+队员标志+形状判别的色标设计方案,队标和队员标志用于获取位姿信息,利用形状判别进一步的确认所获取的小车车号。这种方法减少了颜色的使用种类,避免了某些颜色间的相互干扰,并根据色标的设计,给出了获取目标体位姿信息以及机器人车号的算法。由于广角成像的视觉系统获取的图像存在一定程度的几何畸变,文中在分析了典型的畸变校正模型的基础上,根据足球机器人比赛的要求,提出了适合足球机器人系统的几何畸变校正算法。为了快速、准确识别目标体,本文采用K-均值聚类的图像分割方法和重心法相结合获取目标的中心,以及基于图像获取和目标运动连续性的动态窗口局部搜索算法进行目标辨识。另外,通过编程调试很好地实现了上述所有算法,并且能够满足系统的实时性和精度要求。论文最后介绍了视觉系统软件的各组成模块功能和程序流程,给出了系统运行时的人机交互界面及其具体组成部分介绍,并最终给出了调试结果。另外,文中还指出一些不足之处,为以后的改进提供了一些思路。更多还原

【Abstract】 Soccer robot has become one of the hot subjects in the field of robot researching, it has referred to many kinds of advanced research from different fields, actually it is a research platform which is full of high technologies as challenges. The vision system is an important part in soccer robot system as one of four subsystems, its main task is to capture images through video equipments, and obtain the position and angle information of objects through proper image process, which can provide original information for decision-making system. The speed and precision of recognition are important factors to evaluate the performance of vision system, so the emphasis in this paper is how to improve the recognizing precision and real time of vision system.The SCC-331 camera of SAMSUNG and the image grab card CA-CPE-3000 is adopted firstly to capture images as far as domestic researches. The vision system can obtain superior images with high speed, which is not only easy for later image process, but also can greatly improve the real time of the system.According to the characteristics of new camera selected in this paper, RGB color model is used to process image after analyzing the several color models normally used in vision system, which can reduce the calculation in the system. A new design of color patch is introduced after analyzing the traditional color patch, which can decrease the total kinds of color used in the color patch and avoid the mutual disturbance between some kinds of colors. A set of algorithm is presented to obtain the position and angle information according to the new design of color patch. Because of the characteristics of wide-angle imaging system, geometrical distortion exists in actual images to some extent, a new kind of distortion correction model is built based on the classic model which is fit for the robot soccer competition. To recognize the objects quickly and accurately, image segmentation based on K-means Clustering and center of gravity method, as well as searching in a dynamic window based on the continuity of image capturing and object moving, are adopted to realize the recognition. All the algorithms above are realized well through programming and debugging, it is proved that the algorithms can meet the need of real time and precision greatly for the vision system.In the last, the function of modules in the software and the program flow are presented, man-machine interface is introduce as well as its components and running result of vision system software is provided. In addition, some shortcomings are pointed out and some thoughts are given for the next development.更多还原