【作者】 韩建平；

【导师】 潘志庚；

【作者基本信息】 浙江大学 ， 计算机科学与技术， 2009， 博士

【摘要】 视频运动对象分割是计算机视觉和视频处理领域一项关键技术,具有重要的研究和应用价值。本文针对单目、静止摄像机采集的视频,研究运动对象分割问题,取得的研究成果包括:(1)提出一种基于均值漂移的背景建模及运动对象分割算法。通过均值漂移方法检测概率密度的模式,实现准确的背景估计。针对均值漂移计算复杂度高的局限性,根据帧间时域相关性,提出简化的均值漂移算法;同时通过基于四叉树结构的层次化方法减少逐像素检测造成的冗余计算,先在粗的尺度上搜索运动像素,再逐步以更细的尺度渐近优化运动物体分割。(2)提出一种像素层背景模型及运动对象分割方法。本文利用背景运动局部性和时空变化相关性特征,将背景表示为一组具有相同统计特征的像素层,通过与像素邻域的层匹配来实现运动对象提取。本文在摄像机晃动等原因引起的像素时域变化不规则情况下,具有更稳定的分割效果。算法在空间与时间复杂度方面具有显著的优势。(3)提出一种基于图切分的视频运动对象分割算法。首先建立基于像素层的背景模型,并在对视频帧初步分割的基础上建立前景和阴影模型。采用直方图统计的方法估计阴影对背景像素产生的衰减比例,以建立更准确的阴影模型。通过MRF随机场描述邻域像素间的空域一致性关系,利用图切分方法来求解视频运动对象分割问题。在研究上述算法的基础上,开发了海事场景智能视频监控系统。系统从网络上获取海事场景的视频流,对用户定义的监控区域进行运动目标检测与跟踪,并根据用户定义的规则对异常事件进行告警。通过高效的运动目标分割与跟踪算法,以及分区域监控策略等方式来提高系统的速度,系统在微型计算机平台上实现多路视频的实时自动监控。更多还原

【Abstract】 As a key supporting technique for computer vision,moving object segmentation has far-going pragmatism significance and application importance.This dissertation presents some efforts on extracting moving objects from monocular videos captured by static camera,and main contributions of my work include:(1) Mean shift based background modeling and moving objects segmentation algorithm.Mean shift based non-parametric background modeling supports more sensitive and robust detection in dynamic outdoor scenes.This algorithm aims to deal with the limitation of high computational complexity.Firstly,fast mean shift approach is presented according to temporal dependencies.Secondly,coarse to fine method is proposed to avoid raster scanning entire image.Foreground pixels are detected in coarse level to roughly locate the foreground objects in the image,and then fine detection is performed on the corresponding blocks gradually.(2) A background model based on pixel layer for moving objects segmentation.Fast mean shift approach is used to cluster into layers those pixels that share similar statistics.The background is then modeled as a group of pixel layers.An incoming pixel is detected as foreground if it does not adhere to these layer-models of the background.The proposed method performs better than the tradtional MoG method under temporally irregular dynamic textures.(3) A moving objects segmentation algorithm based on graph cuts.The background model is represented as a group of pixel layers,and the foreground and shadow models are learned from background subtraction.We design a histogram based method to estimate darkening ratio caused by moving shadow so as to model the shadow more accurately.Markov Random Field is used to model the dependencies among neighbouring pixies,and the final foreground segmentation is subsequently achieved by the graph cuts algorithm.We also developed an automatic video surveillance system for marine scenes.It accesses the video streams of marine scenes transferred through Internet and performs moving objects detection and tracking to discover the prohibited objects and alarm acording to user-defined rules.The system supports real time monitoring as many as eight channels of video stream on personal computer.更多还原