导　　师： 郑志强

学科专业： H1101

授予学位： 硕士

作　　者： ;

机构地区： 国防科学技术大学

摘　　要： 本论文以地面自主移动机器人为研究平台,结合RoboCup中型组比赛的实际应用为课题背景,设计并实现了一套由全向反射镜、图像采集设备以及图像处理软件组成的全景视觉系统,用来完成实时图像采集、目标识别和机器人定位等功能。全景视觉系统能够在RoboCup中型组比赛场地上观察到整个场地的图像,根据球门、立柱等标志物进行定位,成功地找到球和双方机器人的位置。整个系统运行稳定,在保证采集速度和图像处理时间的条件下,基本能够满足比赛的实时性和定位精度要求。 论文的主要工作如下:设计并实现了一套适用于RoboCup中型组比赛的全景视觉系统结构,选定摄像机和全向反射镜的形状。通过仿真实验证明双曲线镜面具有单视点、变形小、分辨率高、体积小的特点,能够较好的满足比赛的要求。根据这一结果,设计并加工了双曲线反射镜,获得了较满意的全景图像效果。对用自行设计并实现的全景视觉系统采集到的图像进行处理,选择了对光线变化不敏感的HSV颜色模型进行颜色标定,在此颜色空间上进行图像分割和特征提取,识别机器人、球、球门、立柱和白线等特征。针对根据图像提取的特征进行机器人的自定位、确定双方队员和球的位置的问题,研究了利用特征点定位、利用白线定位、Monte-Carlo定位等几种定位方法,并对各种定位方法的效果和特点进行了比较,确定了特征点定位方法为主要的定位方法。 This thesis introduces the design and implement of a catadioptric vision system for the middle-size league of RoboCup - The World Cup of Soccer Robots. The system is composed of a camera and a mirror generated with a surface of revolution, which has a wide visual field of 360 degrees The system can perform the functions of collecting images, object recognition and locating the robots. The system is designed to observe the whole floor from any point on the floor. Self-Localization algorithms are proposed using the features of goals, corner posts and so on to find the right positions of the ball and robots. The whole system is robust、 accurate and fast, which can /(?/)fill the requisites of the middle-size league of RoboCup.In this thesis, the main work is as follows: the catadioptric vision system is designed and implemented including camera choosing and mirror shape design. Through simulation, the proposed design resuhs in a hyperboloidal mirror with the characters of a single viewpoint、 small distortion、high resolution and a small size, which can fulfill the requisites of the middle-size league of R/(?/)oCup. Tae thesis presents a method for calibration of a color classificatory used for color blob-/(?/)ased image segmentation and object recognition. Because the HSV color model is n/(?/) sensitive to the change of light conditions, the features of robots 、 ball、 goal and white lines are detecte/(?/) on the base of HSV color model, which are important features for robot self-localization. Finally. the different localization methods are given, such as localization using features, lo/(?/)lization asing white lines, monto-carlo localization. Compared with each other, the localization methods effects and qualities can be concluded and localization method using features is /(?/)ose as the primary method.

关 键 词： 机器人世界杯足球赛 全景视觉系统 单视点 双曲线镜面 颜色空间模型 特征提取 目标识别 定位

领　　域： [自动化与计算机技术] [自动化与计算机技术]