未知环境下轮式移动机器人视觉稳定控制方法研究

项目名称： 未知环境下轮式移动机器人视觉稳定控制方法研究

项目编号： No.61075079

项目类型： 面上项目

立项/批准年度： 2011

项目学科： 金属学与金属工艺

项目作者： 王忠立

作者单位： 北京交通大学

项目金额： 10万元

中文摘要： 要实现未知环境下移动机器人的视觉伺服，需要解决两个关键问题，即未知环境下的视觉感知和基于视觉的控制。为实现未知环境的感知，可以把环境分为平面区域（可移动区域）和障碍物区域。基于该假设，实现未知环境下基于视觉的平面信息提取是研究要解决的关键问题之一。 分别对基于立体视觉的和单目图像序列的两类平面提取方法进行了研究。基于立体视觉的方法研究中，提出一种能同时完成平面检测和摄像机-机器人位姿标定的方法。在此基础上，提出基于地平面上两点的机器人位姿估算方法。通过适当选择移动机器人的状态变量，采用Backstepping设计了机器人控制率，误差动力学模型表明设计的控制率具有指数收敛的效果。在基于单目图像序列的平面检测中，通过对空间平面点的图像运动模型建模，提出了基于平面流场的检测方法。 为了实现视觉伺服控制过程中特征点的可靠跟踪，对复杂环境下，考虑尺度、旋转、平移、视角、仿射、光照、模糊等变化，以及部分遮挡等情况下的特征跟踪及匹配稳定性进行了研究，提出了基于单演信号的多尺度、多信息融合的配准策略。 在项目的支持下，参加国内外学术交流会议4人次，发表学术论文7篇，申请国家发明专利2项。

中文关键词： 平面检测；光流场；视觉伺服；轮式移动机器人；特征提取

英文摘要： When the mobile robot moves in the unknown environment with vision sensor, two key issues must be solved, perception of the environment and vision based control. For the perception of the environment, we can assume that the mobile robot always moves on the floor plane or nearly flat area because the floor plane always represents a traversable region. From this point of view, we divide the surrounding environment into two groups, the ground plane and the obstacles. Then how to segment the planar region from the scene is a very important task for the visual perception of mobile robot. Two planar region extraction methods are probed, that is stereo-based and monocular image sequence based. A method based on stereo vision which can complete plane extraction and camera-robot calibration synchronously is proposed. Based on the results, the camera pose can be estimated with only two points. Then the pose of the mobile robot can be estimated with the two points. We translate the kinematics and dynamic of a WMR in Cartesian coordinate to polar coordinate, by using backstepping techniques recursively, a control law is designed ensuring asymptotic stability of the closed-loop system. To extract planar region in monocular image sequence, the image point motion on a space plane is modeled, and the conception of planar flow which describes the image motion pattern is introduced. Based on the planar flow, the distance error function in image space between a point and the plane is defined, which can be used for planar point detection. Robust feature matching is very important for motion estimation during the process of visual servoing control for mobile robot. A matching strategy which efficiently combined the multi-scale feature of monogenic signal and color entropy information is developed under the monogenic signal analysis framework. This leads to establish reliable data association which is robust to the changes in scale, viewpoint, illumination and blur. Supported by the Natural Science Foundation, we attended the international conference for four times, and published 8 papers, two invention patents related to the research topic are applied.

英文关键词： planar detection;optical flow;visual servoing;wheeled mobile robot;feature extraction