【作者】 唐旭东；

【导师】 庞永杰；

【作者基本信息】 哈尔滨工程大学 ， 流体力学， 2011， 博士

【摘要】 近年来,随着资源不断消耗,人口激增,海洋研究和开发逐渐成为人类赖以生存新的发展空间,因此作为海洋探测重要组成部分的水下机器人得到广泛的应用。智能水下机器人技术是水下机器人系统研究的热点领域,开展智能水下机器人基于视觉的目标探测与跟踪技术研究,也是实现水下机器人在恶劣且复杂多变的环境中进行水下侦查、作业的关键技术之一。本文主要目标是研究水下机器人实时水下管道检测与跟踪系统。论文以单目CCD摄像机为主要视觉传感器,利用视觉系统测量方法得到水下管道的导航信息,并在此基础上建立了一个用于水下机器人的水下管道检测与跟踪系统。具体研究内容如下：1.简要回顾了基于视觉的水下机器人水下管道检测与跟踪系统在国内外的研究发展现状,指出了当前利用水下机器人进行水下管道检测与跟踪技术的研究难点和研究新方向。2.按照数据结构的抽象程度,将系统中传递的数据信息分为由低至高6个层次,详细描述了各层次内容。提出了视觉系统设计的总体内容,设计水下机器人管道检测与跟踪系统的体系结构,并对单目视觉系统进行构建和分析。3.在图像处理层,对水下成像进行了简要的分析,并且针对水下图像的特点,以及水下机器人实时性和管道检测的准确性要求,介绍了几种比较简单和高效的改进图象处理算法,来降低水下成像造成影响。4.在图像解释层的,针对水下图像的特点,构造离散情况下具有不变性的矩特征方法。并结合神经网络理论,提出了两种具有全局搜索能力的水下目标识别方法：用于目标识别的免疫遗传神经网络的结构、建模和基于超香肠神经网络的识别学习分类决策机制。为了提高系统的准确性和实时性,采用了基于动态窗口技术和基于Kalman滤波数据关联和状态更新的水下管道检测算法。5.在环境理解层,对摄像机的标定方法、原理进行简单介绍与分析。应用摄像机透视投影成像原理,通过坐标变换推导出结构光视觉传感器的模型,建立水下管道的图像平面坐标系与水底平面坐标间的射影变换关系。6.在决策规划与运动控制层,主要解决了水下机器人运动控制与决策规划的协调合作问题。根据从环境理解层获得的水下管道导航信息、自身运动信息通过一定的算法来产生机器人动作序列,并通过智能运动控制算法计算得到机器人各个自由度执行机构所应该提供的推力,并完成管道检测与跟踪的工作任务。最后,通过仿真和水池试验,对论文提出的算法进行验证,试验结果表明,基于以上算法的程序能够满足跟踪系统的实时性要求,而且针对水下管道检测与跟踪任务,本论文所提方法的有效性和可行性。更多还原

【Abstract】 In recent years, with the substantial consume of energy sources and increase of population, ocean research and exploitation gradually become the new development space of people surviving. So AUV (autonomous underwater vehicle) has been widely applied as an important component of ocean high-tech. The intelligent AUV technique is also a pop research area of AUV system, and vision based recognition and track is one of key technologies of AUV patroling and working in the complexities and uncertainties of underwater environments.The main prupose of this paper is to carry out with a real-time underwater pipeline detecion and track system for AUV. Taking monocular CCD (charge coupled device) camera as a vision sensor, the navigation information of underwater pipelines can be acquired by vision-measuring method. On this basis, an underwater pipeline detection and track system for AUV is constructed. The detailed content as follows:1. Give a brief overview to the development, applications and research status of the vision based underwater pipeline detection and track system for AUV in and out side of the country. Point out the difficites and new research directions in current underwater pipeline detection and track system for AUV research area.2. According to the abstract degree of data structure, the data information transferred in this system can be divided into six hierarchies from low to high. Information in each hierarchy is described. The general content of vision system is proposed and the system architecture of underwater pipeline detection and track system for AUV is designed, also a monocular vision hardware and software system is developed.3. In the layer of image processing, the underwater image-forming are briefly analyzed. According to the analysis of underwater imaging and improve the accuracy and real-time performace of this system, then some simpe and high efficient improved image processing methods are put forward.4. In the layer of explation, according to the underwater imaging, the affine invariants based on region moments are constructed. Based on neural network theory, then two new methods of underwater objects recognition with global searching ability are proposed. Structure and modeling of the immune genetic neural network (IGNN) applied to target recognition. In order to improve the accuracy and real-time performace of this system, the detection method based on dynamic window technology and kalman filter for pipeline datas association are applied.5. In the layer of environmental insight, the methods and foundation of camera calibration are introduced and analyzed. The model of a vision sensor based on structured light is constructed with perspective imaging and coordinatie transform principle.The projective relation between pipelines’image plane coordinate system and submarine plane coordinatie system are constructed.6. In the layer of AUV decision plan and motion control, coordinates the behavior of motion controller and the decision plan are solved. According to the nacigational information of underwater pipeline from layer of environmental insight and motion information of AUV, some motion sequences of AUV are generated. The thrust force of ever degree executing unit also could be computed by intelligent motion controller algorithms to fulfil the task of underwater pipeline detection and track.Finally, the proposed algorithm is validated by the AUV trail simulation and pool experiment. The experimental results showe that the procedure based on the above algorithm can perform the real-time track system, and these methods proposed are feasible and effective for the task of pipeline detection and track.更多还原