基于协同分割的多视频目标提取算法研究
发布时间：2019-02-23 23:51
【摘要】：随着智能手机及平板电脑等电子产品的普及,以及微信、微博等媒体传播平台的快速发展,图像和视频已经越来越深入地影响人类的生活方式。其中,视频作为最丰富的信息载体之一,在各个行业中有着广泛应用。随着视频信息的爆发式增长,如何让计算机理解视频场景内容,并快速有效地提取人们所需的信息等问题变得越来越重要。其中,视频分割作为视频内容分析的基本步骤,对视频信息后期处理起到了关键作用,因此也受到了越来越多研究学者的关注。另外,视频协同分割作为一项更具有挑战性的课题,通过视频间的特征一致性,将视频集合中的共同目标进行分割,极大地改善了单个视频的分割效果。本文主要围绕视频目标分割领域的一些问题进行了研究,采取合适的协同分割模型,建立视频集合间的联系,解决了视频中非相关帧(帧内不包含目标)的干扰问题,以及通过获取具有全局一致性的目标跟踪链对视频协同分割提供可靠的先验信息。具体的研究工作如下:一,对视频分割中所涉及的几种关键技术进行了介绍,详细介绍了后文中所采用的可能目标区域生成方法,显著性提取算法,并分别对几种不同的方法进行了对比分析,详细介绍了基于图像协同的多目标搜索机制及基于光流的运动信息提取方法,为后文提出新的视频协同分割方法提供理论基础;二,提出了基于目标类选取及非相关帧检测的多视频协同分割方法。该方法中提出首先对视频帧生成多样化的可能目标区域候选集,并将其聚类划分为多个目标类别,其次构建基于目标类的图模型,筛选出视频间共同包含的目标,最后采用非相关帧判别机制及图割优化框架,实现非相关帧的筛选及相关帧的目标分割;三,提出了基于目标跟踪链的视频多目标协同分割方法。利用前后帧之间的相关性及全局一致性,对目标区域进行跟踪并获取目标跟踪链,从而得到较为可靠的目标运动轨迹及先验信息,并采用能量优化框架完成最终共同目标的协同分割及多类目标分割。
[Abstract]:With the popularity of electronic products such as smart phones and tablets, as well as the rapid development of WeChat, Weibo and other media communication platforms, images and video have more and more profound impact on the human way of life. Among them, as one of the most abundant information carriers, video is widely used in various industries. With the explosive growth of video information, how to make the computer understand the content of video scene and extract the information that people need quickly and effectively becomes more and more important. As the basic step of video content analysis, video segmentation plays a key role in the post-processing of video information, so it has attracted more and more researchers' attention. In addition, as a more challenging subject, video co-segmentation can greatly improve the segmentation effect of a single video by using the feature consistency of video to segment the common target in the video set. This paper mainly focuses on some problems in video target segmentation field, adopts appropriate cooperative segmentation model, establishes the relationship between video sets, and solves the interference problem of non-correlated frames (no target included in frames). And the target tracking chain with global consistency can provide reliable prior information for video cooperative segmentation. The specific research work is as follows: firstly, several key technologies involved in video segmentation are introduced, and the methods of generating possible target regions and salience extraction algorithm are introduced in detail. Several different methods are compared and analyzed respectively. The mechanism of multi-object search based on image collaboration and the method of motion information extraction based on optical flow are introduced in detail. Secondly, a multi-video cooperative segmentation method based on target class selection and uncorrelated frame detection is proposed. In this method, first of all, the possible target region candidate sets are generated for the video frames, and the candidate sets are clustered into multiple target categories. Secondly, a graph model based on the target class is constructed to screen out the targets that are included among the video frames. Finally, uncorrelated frame selection mechanism and graph cut optimization framework are used to realize the selection of uncorrelated frames and the target segmentation of correlated frames. Thirdly, a video multi-target cooperative segmentation method based on target tracking chain is proposed. By using the correlation and global consistency between the frames, the target region is tracked and the target tracking chain is obtained, so as to obtain more reliable target track and prior information. And the energy optimization framework is used to complete the cooperative segmentation and multi-class target segmentation of the final common goal.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
[Abstract]:With the popularity of electronic products such as smart phones and tablets, as well as the rapid development of WeChat, Weibo and other media communication platforms, images and video have more and more profound impact on the human way of life. Among them, as one of the most abundant information carriers, video is widely used in various industries. With the explosive growth of video information, how to make the computer understand the content of video scene and extract the information that people need quickly and effectively becomes more and more important. As the basic step of video content analysis, video segmentation plays a key role in the post-processing of video information, so it has attracted more and more researchers' attention. In addition, as a more challenging subject, video co-segmentation can greatly improve the segmentation effect of a single video by using the feature consistency of video to segment the common target in the video set. This paper mainly focuses on some problems in video target segmentation field, adopts appropriate cooperative segmentation model, establishes the relationship between video sets, and solves the interference problem of non-correlated frames (no target included in frames). And the target tracking chain with global consistency can provide reliable prior information for video cooperative segmentation. The specific research work is as follows: firstly, several key technologies involved in video segmentation are introduced, and the methods of generating possible target regions and salience extraction algorithm are introduced in detail. Several different methods are compared and analyzed respectively. The mechanism of multi-object search based on image collaboration and the method of motion information extraction based on optical flow are introduced in detail. Secondly, a multi-video cooperative segmentation method based on target class selection and uncorrelated frame detection is proposed. In this method, first of all, the possible target region candidate sets are generated for the video frames, and the candidate sets are clustered into multiple target categories. Secondly, a graph model based on the target class is constructed to screen out the targets that are included among the video frames. Finally, uncorrelated frame selection mechanism and graph cut optimization framework are used to realize the selection of uncorrelated frames and the target segmentation of correlated frames. Thirdly, a video multi-target cooperative segmentation method based on target tracking chain is proposed. By using the correlation and global consistency between the frames, the target region is tracked and the target tracking chain is obtained, so as to obtain more reliable target track and prior information. And the energy optimization framework is used to complete the cooperative segmentation and multi-class target segmentation of the final common goal.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41
【相似文献】
相关期刊论文 前10条
1 靳晓辉;候正信;;智能监控系统中运动目标分割算法[J];电子测量技术;2005年06期
2 徐军，，向健勇，周翔，杨宜禾;基于特征域的目标分割算法[J];红外与激光工程;1998年02期
3 袁天云,姜志国,孟如松;目标分割图中粘连对象的自动切割和分离[J];中国体视学与图像分析;2003年01期
4 李小俊;姚驰甫;;背景重建的运动目标分割改进算法[J];计算机工程与应用;2009年27期
5 邓颖娜;朱虹;刘薇;钱慧芳;;基于姿态模型的人群目标分割方法[J];计算机工程;2010年07期
6 李晶晶;管业鹏;叶勇;;复杂背景下的运动目标分割与阴影消除[J];电子器件;2011年05期
7 李林;卢焕章;肖山竹;廖斌;;基于部分投影的粘连目标分割[J];电视技术;2013年11期
8 侯格贤,吴成柯;一种结合遗传算法的自适应目标分割方法[J];西安电子科技大学学报;1998年02期
9 李艳,彭嘉雄;基于分维特征的目标分割与检测[J];华中理工大学学报;2000年08期
10 杨益军,赵荣椿;一个快速的人机交互式目标分割系统[J];西北工业大学学报;2000年02期
相关会议论文 前5条
1 吴明军;;自由运动像机下的运动目标分割[A];豫赣黑苏鲁五省光学（激光）学会联合学术2012年会论文摘要集[C];2012年
2 袁天云;姜志国;;目标分割图中粘连对象的自动切割和分离[A];图像 仿真 信息技术——第二届联合学术会议论文集[C];2002年
3 刘o
本文编号：2429349
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2429349.html
