基于RoboCup标准足球机器人的目标识别方法研究

发布时间：2018-02-03 16:47

本文关键词： 足球机器人视觉目标识别单目测距级联AdaBoost　出处：《南京理工大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着人类科技水平的提高,智能机器人技术得到空前的发展。RoboCup(机器人足球世界杯)作为机器人领域一项国际性的大型比赛,为在世界各地的众多从事机器人技术研究的科研人员提供了一个广阔的交流平台,推动着机器人技术的进步。在RoboCup中,视觉是机器人感知环境信息的重要来源,在整个系统中起着至关重要的作用。本文以RoboCup足球赛为背景,围绕比赛环境下足球机器人的视觉系统展开研究。首先,视觉系统的基本功能是实现对场地目标的检测识别。本文中分析了NAO机器人的图像获取方式,选取最快捷的方式进行图像采集;针对特殊的比赛环境,选择HSV颜色空间模型,在滤除场外背景之后,实现对目标图像的快速分割,再对分割后的图像进行降噪处理;场地线是球场中的重要标识物,本文提出了一种基于扫描线种子生成的快速场线识别算法,并在此基础上,实现了对场线交叉点、球门柱等场地中特殊标识物的快速精确地检测识别。其次,机器人还应能获取观测物体的深度信息,为机器人运动、定位提供支持。NAO机器人只具有单目视觉功能,为此,本文首先分析了目前几种常用的单目测距模型,然后,以地平面约束模型为基础,推导出了适用于NAO机器人的单目视觉测距模型;对摄像头进行标定实验,获取其内外参数,并基于该测距模型进行单目测距实验,实验结果证明了该模型的有效性和准确性。在足球比赛中,对球员的识别是进行路线规划、传球等上层策略的基础。本文提出将类Haar特征和级联Adaboost算法应用到对NAO的识别中,并利用颜色直方图匹配对候选目标区域进行二次识别,在排除误检区域的同时,也进行敌我识别。实验结果表明,本方法能够有效地识别NAO机器人,满足比赛中识别算法对鲁棒性和实时性的要求。
[Abstract]:With the improvement of human science and technology, intelligent robot technology has been unprecedented development. RoboCup.RoboCupis an international large-scale competition in the field of robotics. It provides a broad communication platform for many researchers engaged in robot technology research all over the world, and promotes the progress of robot technology. In RoboCup. Vision is an important source of environmental information for robots and plays an important role in the whole system. This paper takes RoboCup football as the background. First of all, the basic function of the visual system is to realize the detection and recognition of site targets. In this paper, the image acquisition method of NAO robot is analyzed. Choose the quickest way to collect images; According to the special competition environment, the HSV color space model is selected. After filtering off the field background, the target image can be segmented quickly, and then the segmented image can be de-noised. The field line is an important marker in the field. In this paper, a fast field line recognition algorithm based on scanning line seed generation is proposed, and on this basis, the intersection of the field line is realized. The detection and identification of special markers in the field such as ball doorposts should be rapid and accurate. Secondly, the robot should be able to obtain the depth information of observation objects, which is the motion of the robot. The localization support. NAO robot only has the monocular vision function. Therefore, this paper first analyzes several commonly used single visual distance models, and then based on the ground plane constraint model. The monocular vision ranging model for NAO robot is derived. The camera is calibrated to obtain its internal and external parameters, and based on the distance measurement model, the experimental results show that the model is effective and accurate. The recognition of players is the basis of the upper strategy of route planning and passing. In this paper, the similar Haar features and cascaded Adaboost algorithm are applied to the recognition of NAO. The color histogram matching is used to recognize the candidate target area twice, while the false detection area is eliminated and the foe and I are also identified. The experimental results show that the proposed method can effectively identify the NAO robot. It meets the requirements of robustness and real-time performance of the recognition algorithm.
【学位授予单位】：南京理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【相似文献】