【作者】 曾文静；

【导师】 万磊；

【作者基本信息】 哈尔滨工程大学 ， 船舶与海洋结构物设计制造， 2013， 博士

【摘要】 海洋作为世贸的主要渠道，蕴含着广阔的自然资源。漫长的海岸线也是国家安全的第一道防线。大力发展海洋装备，提高对海洋的控制能力和开发能力，一直受到我国和众多西方国家的重视。水面无人艇（Unmanned Surface Vehicles，USV）作为一种水面无人航行器，它的研究内容涉及到系统工程、自动控制、计算机科学、船舶结构力学、流体动力学等众多学科。因此，对USV的研究有重要的理论意义和工程实用价值。本文以“XL-USV”为研究对象，重点研究了基于可见光视觉传感器的水面目标检测跟踪技术，通过水面图像处理，得出有效信息如水界线以及水面目标或障碍物的位置等，作为USV自主规避障碍物或自主作业的必要条件，提高USV自身的生存与作业能力。具体研究内容如下：（1）回顾了国内外USV的发展现状以及USV结合环境感知设备的技术应用状况，总结了基于光视觉的水面图像处理技术的研究现状，包括水界线检测技术，水面目标检测技术以及水面目标跟踪技术。（2）分析并总结水面图像的典型特征。然后针对水面图像的特点，从图像的质量增强和平滑两个方面，进行了算法研究，提出了改进的增强和平滑处理方法，解决了图像中的曝光和反光等高亮度区域的去除问题。（3）分析了USV在航行过程中的水面图像背景特性，总结了水界线在动态场景图像中的表现形式，提出了一种既适合于水天线提取又适合于水岸线提取的方法，解决了USV在航行过程中对水界线的自适应提取问题。（4）分析了水面目标（障碍物）在背景变化的序列图像中的表现形式，提出了改进的Mean-Shift分割方法；借鉴PhotoShop软件中“浮雕”的图像效果处理原理，提出了基于横纵梯度信息融合的目标提取方法，解决了USV在航行过程中实时准确的目标（障碍物）定位问题。（5）研究了水面目标跟踪方法的实现，通过将经典的跟踪方法Mean-Shift搜索模型和Kalman滤波预测模型用于水面图像中目标位置信息跟踪，融合两者自身的优势，尝试构建了两种预测搜索跟踪框架，改善了跟踪速度，降低了目标尺度变化的影响。（6）介绍了本文研究对象“XL-USV”的体系结构及其环境感知系统的组成；建立了视觉感知系统的仿真子平台，验证了图像处理算法流程以及视觉与规划控制之间的数据通信；与载体进行系统集成，在海试过程中，通过USV的自主避障航行实验，完成了对方法的可靠性与有效性的验证。更多还原

【Abstract】 The ocean is main channel of world trade. There is a great of natural resource. It isalways the first security defence of one country as well. More and more attention has beenpaid to develop marine equipment and improve the ability of ocean exploitation andcontrolling by our country and many west countries. As a kind of surface unmanned vehicle,USV involves vehicle structure, hydromechanics, computer science, robotic vision, automaticcontrol and other areas. Therefore, the research on USV is significant in both theoretical andpractical aspects.Object detection and tracking of marine image sequences based on visual sensor ismainly discussed in this paper, aiming at a certain USV “XL-USV”. Effective informationsuch as sea line and object position can be obtained through image processing and it isnecessary for USV to prevent collision and carry out tasks autonomously. The detailedcontents are as following:(1) Give a brief overview to the development of USV and the applications status of USVbased on environment perception equipment in and out of the country. Summerize theresearch of marine image processing including sea line detection and objection detection andobject tracking technology.(2) Typical feature of marine image is analyzed. Image enhancing methods and imagesmoothing methods are studied combining with those features. A practic method is present torestore the demaged image with noise such as exposures and reflections.(3) The background characteristic of marine image sequences captured during USVmoving is analyzed. The form of sea line displayed in dynamic sence is summarized. Afeasible method is proposed to detect sea-line which is not only appropriate for sea-skybackground but also for offshore background. The problem of moving USV detecting the sealine adaptively is solved.(4) The form of object distribution in marine image sequences is analyzed. TraditionalMean-shift segmentation is improved for this special segmenting condition. Meanwhile, agradient information fusion method is used to locate objects in the light of embossment,which is a tool of Photoshop software. The problem of moving USV locating object quicklyand exactly is solved.(5) Surface object tracking methods is discussed. Classic tracking approach such as Mean-shift searching mode and Kalman filter prediction mode is employed in object trackingof marine images. Advantages of these two methods are taken into account and two frames ofprediction and searching modes is constructed to track objects. They release the problem oftracking velocity and tracking size.(6) The system architecture and environmental perception system of XL-USV isintroduced. The visual perception simulation platform of USV is built. Processing flow istested on the platform. The data communication between visual system and path-planning andcontrolling system is smooth. During sea test, the visual perception system is embedded in thereal USV to help it avoid obstacle and move autonomously. It is validated that the processingflow is robust and effective.更多还原