SIA OpenIR  > 光电信息技术研究室
基于视觉的水下机器人悬停定位方法与应用研究
Alternative TitleVision-based Station Keeping Method and Application of Underwater Vehicles
吴清潇1,2
Department光电信息技术研究室
Thesis Advisor朱枫 ; 郝颖明
ClassificationTP242.3
Keyword水下机器人 计算机视觉 悬停定位 位姿估计 摄像机标定
Call NumberTP242.3/W83/2007
Pages101页
Degree Discipline模式识别与智能系统
Degree Name博士
2007-05-25
Degree Grantor中国科学院沈阳自动化研究所
Place of Conferral沈阳
Abstract为解决受外力扰动而影响定点作业的问题,水下机器人需要具有悬停定位(Station keeping)的功能。悬停定位是指在存在外界扰动的情况下,机器人相对于作业目标仍然能够保持期望的位置和姿态。它具有两个特点:首先,是一种动力定位,即在状态感知系统的引导下,依靠自身动力抵抗外界扰动而保持期望位姿;其二,是一种局部定位,机器人相对作业目标的空间运动范围一般不会很大。因此,对悬停定位的研究需要重点解决两个问题,一个是状态感知问题;另一个是控制问题。 传统的声学定位难以满足精度和实时性的要求,视觉是一种实现水下机器人悬停定位的重要方法。本文以国家“十五”863计划重大专项“7000米载人潜水器”的悬停定位为应用背景,从摄像机标定、特定水下目标识别、视觉伺服和演示实验四个方面进行基于视觉的水下机器人悬停定位方法与应用研究。所开发的系统采用特定观察目标作为合作目标,应用基于模型的单目位姿估计方法获取摄像机与观察目标之间的相对位姿信息,实现了悬停定位中的状态感知;然后以视觉系统提供的位姿信息为反馈,构建视觉伺服控制器,实现了水下机器人的闭环控制。 摄像机标定与水下特定目标识别是实现单目位姿估计的前提,其精度直接影响位姿估计的精度,进而影响整个悬停定位系统的性能。通常摄像机标定是一个先建立成像模型,然后求解模型参数的过程。常用的简化成像模型无法精确描述3维成像空间与2维图像平面之间的关系,并且常规标定方法存在非线性方程优化求解困难的问题。因此,本文提出一种基于虚拟摄像机的无模型标定方法。该方法把摄像机的成像过程作为一个黑盒,通过光电测量方式直接建立3维成像空间与对应2维图像平面之间的映射关系。然后根据该映射关系定义一台虚拟理想摄像机,其透视模型参数可以根据需要任意设定,而不影响最终标定结果。虚拟摄像机的引入使得本方法的应用与常规方法同样方便。实验结果表明,该标定方法可以提高位姿估计精度,特别适用于无法用数学模型精确描述成像过程的系统。 针对系统中应用的特定观察目标,设计了水下图像处理和目标识别算法。对水下图像增强处理以后,应用基于自动阈值和区域生长相结合的方法进行图像分割;然后提取观察目标的图像特征,应用基于模型的目标识别方法实现了水下目标的识别和定位。 以单目位姿估计获得的位姿信息为反馈,构建水下机器人的闭环控制器,以实现悬停定位。这是一个典型的基于位置的视觉伺服问题。针对悬停定位的特点,设计了注视优先原则,在机器人运动过程中合理规划各自由度的控制,应用专家PID控制方法实现了水下机器人视觉伺服。 应用上述研究成果,以水下机器人实验平台为载体,在室内水池搭建了演示实验系统,在国内首次完成了基于视觉的水下机器人悬停定位演示实验。实验结果表明:在悬停定位起始阶段,以观察目标在摄像机视场内为前提,机器人能够在视觉伺服控制下跟踪观察目标,并且使观察目标始终保持在摄像机视场之中;能够运动到指定位置并且保持一个特定姿态,实现了机器人的定位定姿;在受到外界扰动(外力或水流)的情况下,仍然能够恢复到原来的位姿;即使在持续水流的冲击下,也能稳定地保持期望位姿。水下机器人悬停定位演示实验为今后基于视觉的悬停定位技术应用于实际作业打下了良好基础。
Other AbstractIn order to solve the problem of disturbance rejection on fixed-point operations, station keeping has been proposed as a critical capability of underwater vehicles, which requests vehicles can maintain expected pose with respect to a certain object in presence of the disturbance. Station keeping takes on two characteristics. To begin with, it is a dynamic maintenance process, namely, with the guidance of the sensors, vehicles preserve a certain pose by its power under the command of a control system. Secondly, station keeping is a kind of local localization, under consideration that the moving range between underwater vehicles and the operated object is normally not so wide. Summarily, researches on station keeping should cover two crucial issues, sensing and control. Since the conventional acoustic positioning method cannot meet the requirement of precision and real-time performance, vision comes to be one of the most importance methods for station keeping of underwater vehicles. In this dissertation, with the station keeping mission of 7000-meter manned underwater vehicle as the application background, a research on station keeping of underwater vehicles has been developed based on the following four crucial techniques: camera calibration, recognition of special object, visual servoing and experiment test. Treating the observation object as a pre-known model, the relative pose between the camera and the object can be obtained by a pose estimation algorithm using monocular vision, and the sensing problem can be solved therefore. Then the pose information provided by vision system can be fed to visual servoing system of underwater vehicles to solve the control problem of station keeping. Camera calibration and recognition of special object is the premise of pose estimation using model-based monocular vision. The precision of them will influence the estimation accuracy directly and thus influence the performance of the whole station keeping system. Generally, the camera calibration process is divided into two steps: imaging modeling and parameter computation. Unfortunately, the general imaging model cannot precisely describe the relationship between 3D space and 2D image, and even worse, the normal calibration approach suffers from the problem of optimization for the established nonlinear model. To solve the above two problems, a virtual camera calibration methodology without modeling is proposed. The imaging process of camera is treated as a black box and the mapping relationship between 3D object and 2D projection is constructed directly through photoelectric measurement technique. After that a virtual camera can be designed based on the mapping relationship, with randomly set model parameters which have no influence on the final calibration results. The introduction of virtual camera makes our application as convenient as the general methods. Experimental results show that our method can improve the calibration precision, especially for systems whose image process cannot be modeled precisely. The underwater image process and object recognition algorithms are designed with respect to the special observation object. After the image enhancement process, the automatic threshold and region growing techniques are cooperated to realize the image segmentation operations. Then the feature of objective image can be extracted and the recognition and localization of the underwater object can be implemented through model based object recognition approach. Using the pose information from monocular vision as the feedback, a closed loop controller can be constructed to realize station keeping. It is a typical pose based visual servoing architecture. With regard to the trait of station keeping, the principle of watch first is proposed to regulate the control of each degree of freedom, and then the expert PID controller is developed to realize the visual servoing of underwater vehicles. The experiment system of underwater vehicles has been constructed within indoor pool environment, and the first test of station keeping in China is completed based on the above-mentioned researches. It can be concluded from the experimental results that at the initial stage, assumed that the observation object is in camera’s filed of view, the vehicle can track the object and keep watching it all the time; the vehicle can move to any given position and keep desired pose by vision servoing; moreover, the vehicle can restored to original pose in the presence of disturbance (external force or water flow) by itself; even under the persistent water flow, the vehicle can maintain its pose steadily and so as to realize the station keeping. The station keeping demonstrated experiment of underwater vehicles provides pioneer contribution as the preliminary foundation of the future station keeping technology by computer vision in practical applications.
Language中文
Contribution Rank1
Document Type学位论文
Identifierhttp://ir.sia.cn/handle/173321/292
Collection光电信息技术研究室
Affiliation1.中国科学院沈阳自动化研究所
2.中国科学院研究生院
Recommended Citation
GB/T 7714
吴清潇. 基于视觉的水下机器人悬停定位方法与应用研究[D]. 沈阳. 中国科学院沈阳自动化研究所,2007.
Files in This Item:
File Name/Size DocType Version Access License
LWWQX01.pdf(2372KB) 开放获取--Application Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[吴清潇]'s Articles
Baidu academic
Similar articles in Baidu academic
[吴清潇]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[吴清潇]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.