【作者】 丘柳东；

【导师】 李祖枢；

【作者基本信息】 重庆大学 ， 控制理论与控制工程， 2011， 博士

【摘要】 机器人足球是一个新兴的交叉学科,涉及机器人学、智能控制和图像处理、计算机视觉等多个领域,是目前的研究热点与难点。感知系统作为足球机器人的一个重要组成部分,其发展程度对机器人技术的研究有着重要影响。中型组机器人足球比赛(Middle-Size League)是机器人足球世界杯(RoboCup)中的一个重要项目,具有全自主性与极高的实用价值。本文采用中型足球机器人作为研究平台,针对目标识别与自定位中的的全向视觉成像、目标快速识别及其定位、机器人自定位等关键技术进行了研究。首先,在对机器人子系统进行划分的基础上,根据感知系统完成的主要感知功能,构建了机器人感知系统的总体体系结构,对感知系统的层次结构和主要功能的实现进行论述,并指出感知系统需要深入研究的内容。其次,针对全向视觉的主要成像影响因素与球形目标的成像畸变,提出了一种新的全向反射镜面设计方法。全向视觉采用具有针孔投影成像的折反射式,凸型镜面作为基本面型,提出径向长度投影模型,根据球形目标的期望成像构造径向长度投影函数,利用径向长度投影函数,并结合几何光学,实现对镜面形状及参数的求解。再次,针对实时性与精确性要求,提出一种新的快速目标识别及其定位方法。主要由目标区域分割、区域标记、目标区域修复三部分组成。根据目标的颜色特征利用HSI颜色空间对目标区域进行分割;提出了基于轮廓的快速区域标记算法,兼顾实时性与特征获取,获取包括轮廓在内的目标特征;结合目标区域轮廓特征,提出了一种基于凸体的目标修复方法,提高目标的定位精度。最后,针对大比赛场地,提出了一种新的机器人自定位方法。利用全向视觉与白线检测、拟合算法获取白线的初始信息,根据电子罗盘获取的朝向信息,将白线的初始信息转换为白线的半全局信息;根据白线实垂交类型、辅助白线的距离与长度信息,定位区域优先度算法确定机器人所在的最终定位区域;最后根据定位区域的定位特征得到机器人在场地中的位置,实现机器人的自定位。更多还原

【Abstract】 Robot soccer is an emerging interdisciplinary field that involves robotics, intelligent control, image processing, computer vision and other fields. It is currently a hot and difficult research topic. Robot perception system as an important part of soccer robot system, its development remarkably influences the robot technology. Middle-Size League is an important part of RoboCup, It has full autonomy and high practical value. In this paper, used Middle-Size League as a research platform for robot soccer, researched the key technologies of the object recognition and localization, include omni-directional vision imaging modeling, rapid object recognition and its localization, and robot self-localization.Firstly, based on the sub-system dividing of robot, and the main perceptual task, this paper designed the architecture of the robot perceptual system, introduced the hiberarchy of perceptual system and the achievement of perceptual task, point out the contents, which need in-depth research in perceptual system.Secondly, according to the main imaging factor and the distortion of the Spherical object, this paper presented a new omni-directional reflection mirror design method, which used catadioptrics with the pinhole projection as the imaging type and the convex mirror as the basic mirror type. This paper proposed Radial Length Projection Model, constructed Radial Length Projection Function based on the Spherical object Expected imaging, used Radial Length Projection Function and the theory of geometrical optics, obtained the mirror shape and the parameters.Thirdly, according to the real-time and precision requirement, this paper presented a new method for rapid object identification and its localization. This method contains the object component segmentation, the component labelling and the component repair. This paper used HSI color space to segment the object component based on the object color characteristic, presented a rapid component labelling algorithm based on contour, obtain the characteristic include contour, while rapidly processing, used the contour and the character of convex, presented the object component repairing algorithm, improve the precision of object localization.Finally, according to the large game field, this paper presented a new robot self-localization method. This method obtained the initial information of white lines by omni-directional vision, converted the initial information to the part-global information of white lines by the orientation which is obtained through electronic compass, used the while line true vertical intersection type, the length and distance of assistant white line, localization area priority algorithm to identify the best area for localization, obtained the pose by the information of the intersection point, which belong to the true vertical intersection while line, and then achieve robot self-localization.更多还原