移动机器人视觉导航系统研究

北京交通大学

硕士学位论文

移动机器人视觉导航系统研究姓名:王红波

申请学位级别:硕士专业:信号与信息处理指导教师:阮秋琦 20080601

中文摘要

中文摘要

摘要:基于视觉的移动机器人导航是近年发展起来的一种先进导航技术。与其它的非视觉传感器导航相比,它具有信息丰富、探测范围宽、目标信息完整等优点。本文结合实际应用,提出了一个完整的移动机器人视觉导航系统解决方案。研究内容主要包括四个部分:摄像机标定、目标识别、单目测距和运动控制。分别阐述如下:

第一,摄像机标定,基于张正友的平面标定算法对摄像头进行精确标定,针对摄像头的自动变焦特性,提出了一个新的离线离散标定策略,并获得多个状态下的摄像头内外参数。

第二,目标识别,传统分割方法存在多分割问题,影响到目标物提取的精度, 这罩提出一个改进了的基于HSI模型的彩色图像分割算法,在多通道阈值分割的基础上,融入了连通区域标记和形念学开闭运算。

第三,单目测距,基于摄影测量学和立体几何理论,建立了单目视觉测距模型,并推导了基于地平面约束的单目测距算法。针对多种误差因素,在测距算法中加入了误差校币,使移动机器人能够更加准确地定位目标物体。

第四,运动控制,控制摄像机云台实现日标物搜索,调整移动机器人位姿和对夹持器的动作控制。

实验结果表明:即使在恶劣光照条件下,提出的Hs工分割算法能够对向光、背光、近处、远处物体实现快速有效提取;提出的单目测距模型和算法能够对目标物体进行精确的测距;当把这些算法集成到实验平台上时,能够快速实现移动机器人的导航控制,并成功完成物体抓取操作。

关键词:摄像机标定、彩色目标识别、单目视觉测距、移动机器人

分类号:TP 391.41

ABSTRACI'

ABSTRACT

ABS。I’RAC’1..

In recent years,vision attracts a lot of attention for navigating a mobile robot in

dynamic pared with other sensing systems,visual navigation is excellent and effective.With a visual sensing system,wider view of field,rich and

intensive data Can be obtained for a mobile robot moving in a changing environment.In this study,a visual navigation scheme is proposed for a mobile robot to realize object collection,and it comprises of camera calibration,object

recognition,monocular measurement and motion control,as stated in the following.

Firstly,the technique of camera calibration is presented on the basis of Zhang’S algorithm.Since a PTZ calTlera is used here,it is controlled to move up

and down,from left to right,to extend the view of field.Therefore,calibration in different positions is needed,and a new discrete method is proposed here.

Secondly,a

modified image segmentation method is used here based on HS[ algorithm,with the morphological filtering

and multi—channel segmentation techniques・ It improves the accuracy of the conventional segmentation methods.

Thirdly,a monocular technique is proposed here to measure

an object position in 3D space.Visual models are established based on the principle of imaging and spatial geometry constrains.To

reduce the measuring errors,a compensating term is added into the method.

Finally,these techniques

are integrated into a mobile robot platform to implement collection of specified objects.To find an object of long distance,the camera is tilted and panned,the location information is computed from the feedback of wheel motions, and the gripper is control to grasp the specified object.

Numerous experiments show that the calibration of camera is precise,and the image segmentation algorithm is effective even under very bad illumination conditions, and the visual navigation scheme works very well for the mobile robot to collect a specified object.

KEYWORDS:Objeet Recognition,Monocular Measurement,Mobile Robots, Visual Navigation

CI.ASSNo:TP 391.41

VIl

学位论文版权使用授权书