【作者】 周静；

【导师】 文灏；

【作者基本信息】 华中科技大学 ， 通信与信息系统， 2007， 硕士

【摘要】 运动目标的检测与跟踪是计算机视觉和图像编码研究的主要内容,也是近年来计算机视觉中最具挑战性的研究课题之一,它在机器人导航、智能监视系统、医学图像分析以及视频图像压缩和传输等领域中都有广泛的应用。本文从实用性出发,对基于序列图像的背景模型建立与运动物体检测的相关理论进行了研究。本文首先介绍了国内外运动物体检测的各种算法,包括基于差分、基于背景建模、以及基于光流场的算法,分析了各种算法的优缺点,以及应用场景。针对基于差分的背景模型建立算法对于运动物体速度变化时更新不及时的问题,采用了一种新颖的使用对称差分算法进行背景建模的方法。即利用对称差分能快速检测出中间帧运动目标的形状轮廓的优点,使用间隔多帧的对称差分进行背景模型的动态更新,得到可靠的背景模型,并通过调整间隔的帧数来实现环境变化的可适应性。另外,为了提高运动物体检测的正确性,本文采用对称差分信息作为背景消减的有效补充,用于运动目标的检测。在建立了可靠的背景模型的基础上,对比基于差分的运动检测,基于对称差分的算法不仅全面的涵盖了前者的检测范围,同时还能够有效的分离出背景消减无法检测出的区域。实验表明该算法适用于摄像机静止物体运动的场景,并能有效地检测出运动目标,且算法复杂度低,正确性较高,具有一定的实际应用价值。同时本文研究采用标准Hall测试序列进行验证和算法对比分析,体现了研究的普遍意义。更多还原

【Abstract】 The detection and tracking moving targets in serial images are not only the main research field in computer vision and image coding,but also one of the most challenging research field in computer vision recently. It has vast applications in automation navigation, video surveillance and monitoring, medicine image analysis and video image compression and transmission. The research of the paper is focused on the serial image based background modeling and object detection.The paper introduces common approaches and algorithms of object detection home and abroad including the difference method, background modeling method and optical flow method, and then analysis the applicable field, advantages and disadvantages for each algorithm. Since difference based background modeling works poorly when the object speed is changeable, symmetrical difference method based background modeling is proposed. The symmetrical difference can quickly detect the object shape, gain more reliable background model and is more adaptable to the changeable environment.The paper has also proposed a new object detection method to get higher detection correctness, adopting symmetrical difference method into object detection. It totally covers the object detection based on difference, and also efficiently detects those areas in which it works poorly. Efficient and good segmentation performance is demonstrated by the experiment results. At the same time, the paper makes the research work more practical for it adopts the standard Hall serial image for experiments and algorithm analysis.更多还原