用户名: 密码: 验证码:
基于变化区域检测的视频对象分割
详细信息    本馆镜像全文|  推荐本文 |  |   获取CNKI官网全文
摘要
视频对象分割就是把视频按照一定标准分割成区域,目的是为了从图像
    序列中标记出有意义的实体,这些有意义的实体,在视频中称为视频对象。
    通常,一个视频场景由视频对象(运动对象)和背景组成,人可以轻易的从
    视频中分辨出这些部分,如果用计算机来做就很困难。视频中的运动目标是
    视频的重要组成部分,能够分离出这些部分,在很多领域有着重要作用。
    在智能监控的应用中,需要实现交通路口、高速公路、机关门卫过往车
    辆的实时登记和流量统计,这些应用,需要建立在车辆独立分割的前提下;
    在网络传输中,如果把兴趣对象(视频中一般为运动对象)从需要传输的帧
    序列中提取出来,分配更多的资源,当信道质量下降时,虽然画面质量跟随
    下降,但是人们的兴趣中心得以保障,提高了视频的主观质量;在军事领域
    中,图像往往是重要信息的载体,通过连续拍摄的图像构成序列,从中分离
    出运动物体能够给侦察提供线索。目前,对红外序列图像的研究以及通过对
    目标的自动跟踪能够为自动武器的发展奠定基础;在视频编码中,基于内容
    的编码标准 MPEG-4 已经于 1999 年公布,MPEG-4 采用了基于对象和模
    型的编码方法,提高了压缩效率,更加合理的分配和利用传输资源。基于对
    象编码,首先就要对视频对象分割,这个工具并没有包含在标准中,作为标
    准的开放部分留待进一步研究。
    就目前的计算能力和图像的复杂性,理想的在任意场景下分离所需对象
    几乎是不可实现的。针对不同的实际应用有很多适用的方案,但都有适用范
    围,比较常见的处理方法是根据应用特点,利用视频的一种或者几种特性,
    结合不同方法的适用范围选择算法实现预定目标。
    交通领域视频特点是,运动目标为刚体,运动轨迹大多趋于直线。目前
    在高密度场景的分割很难实现,或效果很难让人满意。门禁、高速公路的监
    控中,一般目标数目较少,移动速度快。本文的研究目标是针对交通领域、
    智能监控应用的假设,在摄像机镜头没有移动、变焦等变化,图像背景相对
    简单、稳定时,单一运动目标乃至于具有相同特征的多运动目标的分割提取。
     i
    
    
    吉林大学硕士学位论文
    在综述目前现存的视频对象分割方案的前提下发现,光流估算的方法运算量
    大,而且本文研究背景不符合光照不变假设,因此并不适用本文情况。基于
    块的匹配方法使用灰度作为匹配准则的数据来源,在光照存在突变的区域,
    基于块的匹配方法也无法正确跟踪目标。本文采用变化区域检测的方法寻找
    目标初始位置,然后应用基于变形模板的跟踪方法获取目标最终边界。
     对于慢速运动物体,可以使用帧间差分的方法提取变化区域,然后进行
    跟踪修正。但目标运动速度较快的时候,帧间差分的误差很大,差分图像很
    难体现目标的形状及运动信息。这种情况,可以使用运动目标与背景的差分
    (减背景法)来检测运动区域。在无法适时直接获取无目标背景,或背景缓
    变的情况下,背景重建也不得不加入到视频分割的步骤中来。
     文中通过分析目前常用的背景重建方法,并针对智能监控应用的需求,
    提出了更加适合的背景重建方案——基于距离测度的去除干扰法。通过实验
    比较,Kalman 渐消滤波法基于渐消原理,获得背景后,当运动目标再次扰动
    时,需要一段时间来消除影响,体现在重建的背景上就是目标后比较长的“拖
    尾”;基于高阶统计量的分块方法,对于不同的场景,存在恢复失败的可能,
    致使系统不够鲁棒;本文方法需要较少帧来恢复背景,不同于现存方法的是,
    本文方法不易受运动目标扰动的影响,不会出现重建失败,而且克服了“块
    效应”,是适合本文研究背景的重建方法。
     基于距离测度的去除干扰法的原理是:对于单个象素而言,运动目标所
    在区域的象素可以看作是背景象素上的短时扰动。可以把n帧图像的同一位
    置象素作为一维输入信号做统计,对于偏离中心大的点,判定为车辆的扰动,
    否则认定为背景象素。在得到n帧图像平均的基础上,依照小流量、高速运
    动假设,每象素的均值可认为是趋于背景灰度。然后计算每帧图像象素的中
    心距,中心距大过某预定值,既标定为运动目标象素,恢复背景时去除这些
    象素的干扰。然后求得更趋近真值的背景灰度。
     图像所提供的仅仅是灰度、色彩、纹理、边缘等低等次信息。通常,视
    频中运动目标带有多种灰度、色彩和纹理信息,通过计算来从中恢复高层次
    的语义信息比较困难,因此,根据运动特征,运动一致物体的边缘往往能够
    带来更全面的信息。在众多的分割方法中,Snake 模型具有开放性,用参数
    可调的方法,动态逼进特征边缘,是一种局部边缘提取的重要方法,因此在
     ii
    
    
    摘要
    图像处理、运动跟踪等方面得到广泛的应用。GVF Snake 模型由 Chenyang Xu
    在 1997 年提出,它保留 Snake 模型的内部力,通过改进外部能量,使 Snake
    模型有更大的搜索范围,并且依据图像力判定收缩或者膨胀,不需要事先指
    定 Snake 的运动方向。同时,解决了 Snake 模型必须初始化在期望边界附近
    的缺点。
     本文仿真实验首先通过去除干扰的方法重建序列图像的背景,应用减背
    景法获得
Video object segmentation is to segment video according to certain
    standard in order to mark the important objects, which are called video object.
    General speaking, a video scene is composed of video objects (moving object)
    and it’s background. It is easy for us to distinguish them, but it is difficult for
    computer to do this. It will be a great merit in many fields if these moving
    objects can be divided from the background, for these objects are significant
    parts in video.
     In intelligent surveillance, the vehicles passing from the traffic across,
    highway and guard need to be registered and accounted, which could be
    accomplished only when the vehicles are segmented firstly. In internet transfer,
    if the moving objects can be divided and assigned more resources, when the
    quality of the picture will decrease, observers interest center can be ensured so
    that the subjective quality of video can be increased. In military field, pictures
    are carriers of important information. Taking successive pictures and dividing
    moving objects from it can provide clue for scout. At the present, the research
    of infrared sequential images and automatic tracking of objects have laid a
    foundation for the development of automatic armaments. In video coding, the
    code standard MPEG-4 based on content was publicized in 1994.MPEG-4
    employed the code method based on objects and models, which can increase
    compression efficiency and assign transfer resources more reasonably. Coding
    method based on objects requires the video object to be segmented in the first
    place. The instrument is not included in the standard and waits to be studied
    further.
     Segmenting objects under any condition ideally can hardly be realized due
    to the present capability of calculation and the complexity of images. Each
    project has its own applicability. The usual practice is to choose applicable
     iv
    
    
    method according to different application characteristics, and making use of
    certain characteristics of video.
     The video characteristic of traffic field is that moving objects are rigid
    bodies and the moving tracks are almost straight so that the segmentation in
    high density scenes can hardly be realized satisfactorily. Usually, in the
    surveillance of guard and highway, the number of objects is few and the speeds
    are high. This paper intends to study in the field of traffic intelligent
    surveillance how to segment single moving object or multi-objects of the same
    characteristics in a comparatively simple and stable background when the
    camera doesn’t move or change focus. When analyzing the existing projects of
    video segmentation, it can be found that optical-flow estimation cannot be
    applied in this paper for it demands a large amount of calculation and the
    research background of this paper isn’t in accordance with the hypothesis of
    light invariableness. Since the method of matching block regards gray-level as
    the data resources of matching principle, in the region where light breaks, this
    method cannot follow object correctly. This paper searches the original position
    of objects by the means of varying region detection, and then gets the accurate
    boundary of objects through the tracking method based on distortion template.
     In regard to the low-speed moving objects, variation regions can be
    extracted by the difference between two frames, and then be traced and
    modified. But for the high-speed moving objects, the errors of difference
    between two frames will be great, and the difference image can hardly reflect
    the figure and movement information. Under this circumstance, the difference
    between frame and background can be employed to detect the movement
    region. When the no-object background cannot be achieved directly in time, or
    the background is changing slowly, background rebuilding will have to be
    added into steps of video segmentation.
     Through analyzing the common methods of background rebuilding, and
    taking the requirement of intelligent surveillance
引文
基础文献:
    [1] 杨枝灵, 王开. Visual C++数字图像获取 处理及实践应用. 北京:人民邮
     电出版社, 2003.1: 461-462
    [2] Yao Wang, J?rn Ostermann, Ya-Qin Zhang. 视频处理与通信. 北京: 电子
     工业出版社. 2003.6: 84-151
    [3] A. MURAT TEKALP. 数字视频处理. 北京: 电子工业出版社. 1998.7
    [4] 王宏禹. 现代谱估计. 南京: 东南大学出版社, 1991
    [5] 李在铭. 数字图像处理 压缩与识别技术. 成都: 电子科技大学出版社,
     2000.11
    [6] 张贤达. 现代信号处理. 北京: 清华大学出版社, 1994
    [7] 章毓晋. 图象分割. 北京: 科学出版社, 2001
    [8] Rafael C. Gonzalez, Richard E. Woods. 数字图像处理(第二版). 北京: 电
     子工业出版社, 2003.3
    综述文献:
    [9] 季白杨, 陈纯, 钱英. 视频分割技术的发展. 计算机研究与发展, 2001,
     38(1): 36-42
    [10]刘党辉, 沈兰荪. 视频运动对象分割技术的研究. 电路与系统学报, 2002,
     7(3): 77-85
    [11]韩军, 熊璋, 孙文彦, 龚声蓉. 分割视频运动对象的研究. 计算机工程与
     应用, 2000.8: 22-26
    [12]邬正平. 视频对象分割技术的研究. 浙江大学, 2002.3
    [13]王亮, 胡卫明, 谭铁牛. 人运动的视觉分析综述. 计算机学报, 2002, 25(3),
     225-237
    [14]Murray D W, Buxton B F. Scene segmentation form visual motion using
     global optimization. IEEE Trans on Pattern Analysis and Machine
     56
    
    
    参考文献
     Intelligence, 1987, 9(2): 220-228
    [15]Hotter M, Thoma R. Image segmentation based on object oriented mapping
     parameter estimation. Signal Processing, 1988, 15(3): 315-334
    [16]Musmann H G, Hotter M, Ostermann J. Object-oriented analysis-synthesis
     coding of moving images. Signal Processing: Image Communication, 1989,
     1(2): 117-138
    [17]Diehl N. Object-oriented motion estimation and segmentation in image
     sequences. Signal Processing: Image Communication, 1991, 3(1): 23-56
    [18]Choi J G, Lee S W, Kim S D. Spatiotemporal video segmentation using a
     joint similarity measure. IEEE Trans on Circuits Systems Video Technology,
     1997, 7(2): 279-286
    动态模型文献:
    [19]李军, 周敬利, 余胜生, 陈加忠, 何小诚. 基于块运动矢量加权的 Snakes
     模型及在序列图像分割中的应用. 计算机研究与发展, 2003, 40(2):
     277-282
    [20]黄波, 杨勇, 王桥等. 一种用于视频分割的快速运动估计方法. 电路与系
     统学报. 2001.3, 6(1):69-71
    [21]谢存, 孔祥维. 利用光流预测 Snakes 模型初始轮廓的动态图像轮廓追踪
     方法. 中国图象图形学报, 2000, 5(8): 666-669
    [22]M Kass, A Witkin, K Terzopoulos. Snakes: Active contour models.
     International Journal of Computer Vision, 1987, 1(4): 321-331
    [23]C. Xu, J.L. Prince. Gradient Vector Flow: A New External Force for Snakes.
     Proc. IEEE Conf. on Comp. Vis. Patt. Recog. (CVPR), 1997.7, Los Alamitos:
     Comp. Soc. Press, 66-71
    [24]C. Xu, J. L. Prince. Snakes, Shapes, and Gradient Vector Flow. IEEE
     Transactions on Image Processing, 1998.3, 7(3): 359-369
    [25]C. Xu, J. L. Prince. Generalized Gradient Vector Flow External Forces for
     Active Contours. Signal Processing--An International Journal, 1998.12, 71(2):
     131-139
     57
    
    
    吉林大学硕士学位论文
    [26]C. Xu, J. L. Prince, Global Optimality of Gradient Vector Flow. Proc. of 34th
     Annual Conference on Information Sciences and Systems (CISS'00), 2000.3,
     Princeton University
    [27]L. D. Cohen. On active contour models and balloons. CVGIP: Image
     Understanding, 1991.3, 53(2): 211–218
    [28]L. D. Cohen, I. Cohen. Finite-element methods for active contour models and
     balloons for 2-D and 3-D images. IEEE Trans. on Pattern Anal. Machine
     Intell., 1993.11, 15(11):1131-1147
    [29]R. Courant, D. Hilbert. Methods of Mathematical Physics (vol. 1). New York:
     Interscience, 1993
    [30]St éphanie Jehan-Besson, Michel Barlaud, Gilles Aubert. Video Object
     Segmentation Using Eulerian Region-Based Active Contours. International
     Conference in Computer Vision, Vancouver, Canada, juillet, 2001
    [31]Segen J, Pingali S. A camera-based system for tracking people in real time.
     Proc International Conference on Pattern recognition, Vienna, 1996, 63-67
    背景重建文献:
    [32]刘永信, 魏平, 候朝桢. 视频图像中运动目标检测的快速方法. 仪器仪表
     学报, 2002.10, 23(5): 163-176
    [33]韩光文, 系统辨识. 武汉:华中理工大学出版社, 1988
    [34]王相森. 基于视频的实时车辆识别和分类的改进方法. 微处理机, 2003.2,
     17-19
    [35]张文涛, 李晓峰, 李在铭. 高速密集视频目标场景下的运动分析. 电子学
     报, 2000.10, 28(10): 114-117
    [36]Roland Mech, Michael Wollborn. A noise robust method for 2D shape
     estimation of moving objects in video sequences considering a moving
     camera. Signal Processing. 1998, 66: 203-217
    [37]McKenna Setal. Tracking groups of people. Computer Vision and Image
     Understanding, 2000, 80 (1): 42-56
    [38]Karmann K, Brandt A. Moving object recognition using an adaptive
     58
    
    
    参考文献
     background memory. In: V Cappellini, Time-varying Image Processing and
     Moving Object Recognition. 2. Elsevier, Amsterdam, The Netherlands, 1990
    [39]Kilger M. A shadow handler in a video-based real-time traffic monitoring
     system. Proc IEEE Workshop on Applications of Computer Vision, Palm
     Springs, CA, 1992, 1060-1066
    [40]Stauffer C, Grimson W. Adaptive background mixture models for real-time
     tracking. Proc IEEE Conference on Computer Vision and Pattern Recognition,
     Fort Collins, Colorado, 1999,2: 246-252
    综合分割方案
    [41]贾振堂, 贺贵明, 韩艳芳. 运动视频对象分割的一种快速算法. 中国图象
     图形学报, 2002, 7A(11): 1123-1127
    [42]邓航, 余松煜. 综合利用通用霍夫变换与 Snake 算法对序列图像的分割.
     红外与激光工程, 2004.4, 29(2): 9-25
    [43]韩军, 熊璋, 孙文彦, 龚声蓉. 自动分割及跟踪视频运动对象的一种实现
     方法. 中国图象图形学报, 2001.8, 6A(8): 733-737
    [44]陈朝阳 张桂林. 基于图像对称差分运算的运动小目标检测方法. 华中理
     工大学学报, 1998, 26(9): 34-35
    [45]Hotter M, Thoma R. Image segmentation based on object oriented mapping
     parameter estimation. Signal Processing. 1988, 15(3):315-334
    [46]Laurenz Wiskott. Segmentation from motion: combining Gabor and
     Mallat-wavelets to overcome the aperture and correspondence problems.
     Pattern recognition, 1999,32:1751-1766
    [47]Margues F, Molina C. Object tracking for content-based functionalities. SPIE
     Visual Communication Image Processing, Visual Communication and Image
     Processing. San Jose, 1997, 190-198
    [48]Luis Salgado, Narciso Garcia, Jos é M.Menéndez, Enrique Rendón.
     Dynamic Object Segmentation for Outdoor Analysis. SPIE Visual
     Communication Image Processing, 2002, 4671: 593-603
    [49]SHI Li, ZHANG Zhao yang, WANG Hui. Automatic segmentation of video
     59
    
    
    吉林大学硕士学位论文
     object plane based on object tracking and matching. SPIE Visual
     Communication Image Processing, 2001, 4550: 28-33
    其他文献
    [50]岑峰,戚飞虎,陈茂林. 长期视频监控系统的多分布模型背景差方法. 红外
     与毫米波学报, 2002.2, 21(1): 59-63
    [51]徐晓夏, 陈泉林. 视频交通监控系统. 现代电子技术, 2003.8: 72-77
    [52]刘李杰, 蔡德钧, 翁南钐. 一种面向运动的视频对象分割算法. 计算机学
     报, 2000.12, 23(12): 1326-1331
    [53]翁南钐, 蔡德钧. 视频对象分割与两种面向对象的视频编码器. 电子学报,
     2000.10, 28(10): 106-110
    [54]马丽红, 张宇, 邓健平. 基于形态开闭滤波二值标记和纹理特征合并的分
     水岭算法. 中国图象图形学报, 2003.1, 8a(1): 77-83
    [55]史立, 张兆扬, 马然. 基于运动跟踪匹配技术的视频对象提取. 通信学报,
     2001.11, 22(11): 77-85
    [56]宋立锋, 韦岗, 王群生. 基于模板匹配的视频对象分割. 电子学报, 2002.7,
     30(7), 1075-1078
    [57]贾振堂, 李生平, 贺贵明, 田惠. 一种基于运动边缘检测的视频对象分割
     新算法. 计算机研究与发展, 2003.5, 40(5): 684-689
    [58]黄波, 杨勇, 王桥, 吴乐南. 一种基于时空联合的视频分割算法. 电子学
     报, 2001.11, 29(11): 1491-1494
    [59]马波, 张田文, 李培华. 基于改进外部力的多尺度蛇算法. 计算机学报,
     2003.3, 26(5): 563-568
    [60]夏利民, 谷士文. 基于活动轮廓的运动目标的动态分割. 中国图象图形学
     报, 1999.8, .4a(8): 631-634
    [61]Zeng Zhi-Hong. Lane Detecting and Car Tracking on the Highway. 自动化
     学报, 2003.3, 29(3): 450-456
    [62]Kanade T. Picture processing by computer complex and recognition of human
     faces. Kyoto: Kyoto University, 1973

© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号

地址:北京市海淀区学院路29号 邮编:100083

电话:办公室:(+86 10)66554848;文献借阅、咨询服务、科技查新:66554700