【作者】 曹辉；

【导师】 张均东；

【作者基本信息】 大连海事大学 ， 轮机工程， 2008， 博士

【摘要】 针对船舶运行环境的特殊性和船舶监控管理的特殊需求,本文研究的主要目的是实现对船载成像系统中波动视频的稳像处理以及在此基础上实现的对感兴趣目标的自动跟踪。由于,在船舶监控过程中,通过对采集到的大量视频图像进行分析后,发现各帧图像信息中隐含着一个很重要的共同特征:远景中的水平特征线总是出现在图像中的某一区域附近,如果远处有目标船只出现,则会将其截断。因此,本文充分利用这一特点,找到特征线与视频稳像和目标跟踪之间的关系,并利用相关的数学工具和视频处理方法,在完成视频稳像处理的同时实现对特征线附近目标的跟踪。在对视频的稳像过程中,首先通过对船载成像系统相关模型的分析得出视频帧间、帧内模糊的原因,推导出特征线参数变化与船体运动之间的关系。然后采用小波变换和Hough变换相结合的方式对特征线进行识别、提取,再参照直线方程与全局坐标系的关系,实现对波动视频序列的运动估计。最后通过自适应平滑滤波将船载视频图像的运动分为主观扫描运动、客观扫描运动和高频随机运动三种形式,主观扫描运动通过滤除加以保护,高频随机运动和客观扫描运动通过图像的几何变换和云台反馈分别进行补偿,最终完成视频的双重补偿稳像。在对目标的跟踪过程中,采用了一种用参考点来代替目标物体的跟踪方法。首先,按照图像空间、参数空间、线段空间之间的对应转换关系,根据Hough变换的结果,对特征线上的点进行动态分组、端点融合,从中提取出线段以及线段的端点,并进一步求出截断中心点(参考点)在图像空间中的坐标。然后,通过离散Kalman滤波器对截断中心点的运动轨迹进行估计,计算出各帧中参考点的坐标。最后,利用云台的反馈使运动目标始终处于预定的警戒范围之内,完成对目标的自动跟踪。从工程应用的角度来看,视频稳像处理是目标跟踪实现的前提和基础,而目标跟踪则是稳像处理的应用和目的。本文结合船舶监控环境的特点,利用对特征线识别和提取,将视频稳像和目标跟踪进行了有机地结合。更多还原

【Abstract】 According to the particularity of ship operating environment and the special requirement of ship monitoring and management, the main purpose of this research is to accomplish image stabilizing for fluctuant shipborne video and finish interesting target tracking based on the former. In the process of ship monitoring, form a host ship’s view to observe outside objects, a very important common characteristic hidden in every frame can be found that the horizontal line information always appears in some area and shakes continuously due to the sway of host ship. If some ship target appears in the distance, the horizontal line will be separated into two line segments. So, by taking full advantage of this feature, the relation between video image stabilization and video target tracking was found out. Then, make the use of relevant mathematical tools and video processing method to make video stabilizing, and at the same time to achieve target tracking nearby horizontal feature line.In the process of video image stabilization, after analyzing relative models about shipborne video-image system to obtain the reasons for interframe blur and insideframe blur, and deduce the relation between parameter variation of feature line and ship motion. Then, adopt the combination of wavelet transform and Hough transform to complete identification and extraction of feature line and consult the relation between feature line equation and global coordinate system to finish motion estimation of fluctuant video sequence. According to motion estimation result, adopt self-adaptive smooth filter to make the shipborne video image motion divided into three types, i.e., subjective scanning motion, objective scanning motion and high frequency random motion. The first motion needs to be protected by filtering out, while the other two types need to be compensated separately. So, use image geometry transform to compensate the high frequency random motion, and use feedback of video platform to compensate the objective scanning motion. By the two kinds of compensation for video image motion, video stabilizing is finished, which is so called dual compensation video image stabilization.In the process of video target tracking, a method that uses a reference point to replace the target object for tracking is adopted. Firstly, according to the corresponding conversion relations between image space, parameter space and line segment space, use the Hough transform to realize dynamic grouping and endpoint integration for the points of feature line, and extract the segments and its’ endpoints, calculate the conference point’s coordinate in image space. Then, use discrete Kalman filter to estimate the conference point’s trajectory and calculate its position in every frame. Finally, use the feedback of video platform to make the moving target always be within the alert area, which is set in advance, and finish the automatic tracking of target.In the view of engineering application, the processing of video image stabilization is a premise and basis for the realization of target tracking, and target tracking is an application and purpose for video image stabilization. According to the feature of ship monitoring environment, this thesis makes a well combination for video image stabilization and target tracking by the processing of identification and extraction for feature line.更多还原