【作者】 杨绍武；

【导师】 郑志强；

【作者基本信息】 国防科学技术大学 ， 控制科学与工程， 2009， 硕士

【摘要】 本文针对仿人机器人打乒乓球的任务,设计并实现了基于双目视觉的机器人外部视觉系统和本体视觉系统对快速运动的乒乓球目标进行识别、定位和跟踪,实时地为机器人提供乒乓球在三维空间中准确的位置信息,为实现仿人机器人打乒乓球奠定了基础。本文首先设计了基于双目视觉的仿人机器人视觉系统的硬件和软件结构,以及同步控制方式,对视觉系统的基线距和时间同步误差对测量精度的影响进行了分析,并对乒乓球目标的准确、实时的识别进行研究。利用颜色查找表分类方法对图像进行颜色分类,设计了快速的区域搜索算法在颜色分类结果的基础上获取乒乓球的目标区域,利用Canny边缘检测方法在目标区域中提取边缘信息,然后设计快速CHT算法准确计算出乒乓球目标中心的图像坐标。然后,对乒乓球目标的定位和跟踪问题进行研究。根据摄像机的几何模型和其内、外参数实现对乒乓球目标的定位。并运用了流体力学的方法对乒乓球飞行过程所受阻力进行了分析,在此基础之上建立乒乓球的运动模型,利用卡尔曼滤波方法实现了对乒乓球目标的跟踪,以获得图像中的感兴趣区域(ROI),增强视觉系统对乒乓球识别的实时性和鲁棒性。最后,提出了基于特征区域角点的摄像机外参实时标定方法,使得本体视觉系统能够在线精确地获得其摄像机的外部参数,实现了在其跟随仿人机器人运动时对乒乓球目标位置信息实时、准确的获取。更多还原

【Abstract】 This dissertation designs and realizes an external vision system and an on-board vision system based on binocular vision to recognize, locate and track the fast moving ping-pong aiming at the task of playing ping-pong by the humanoid robot. The precise three dimensional localization information of the ping-pong is provided to the humanoid robot by the vision systems in real-time, which lays the foundation for successfully playing ping-pong by the robot.Firstly, the dissertation designs the framework and the synchronization method of the hardware and software of the vision systems of the humanoid robot based on binocular vision. And the effects caused by the distance of the baseline and the time error of the synchronization are analyzed. The precise and real-time recognition of the ping-pong is also studied in this thesis. The color look-up table method is used to classify the color image. Then a fast area searching algorithm is designed to obtain the goal area of the ping-pong based on the color classification result. In the goal area, Canny edge detector is used to detect the edge information, and then a fast CHT algorithm is designed to calculate the image coordinate of the center of the ping-pong accurately.Secondly, the localization and tracking problems are also studied in this dissertation. The localization of the ping-pong is realized after the geometric camera model and the camera intrinsic and extrinsic parameters are known. And the air resistance to the ping-pong during the flying of it is analyzed using the tools of hydromechanics. Then the motion model of the ping-pong is founded to apply Kalman Filter to track the ping-pong. The region of interest (ROI) can be obtained with the tracking of ping-pong, which improves the real-time and robust performance of vision systems in recognizing the ping-pong.Finally, the dissertation proposed a real-time calibration method for camera extrinsic parameters based on corners of characteristic areas, which qualifies the on-board vision system to obtain the camera extrinsic parameters of it on-line in high precision. Then the precise three dimensional localization information of the ping-pong is obtained by the on-board vision system in real-time even when the vision system is moving with the humanoid robot.更多还原