【作者】 侯伟；

【导师】 卢炎麟；

【作者基本信息】 浙江工业大学 ， 机械设计与理论， 2008， 硕士

【摘要】 在信息化的社会,视频内容因提供了最为丰富和全面的信息要素日益受到现代交通、网络媒体及计算机视觉等行业的青睐和重视。然而,原始视频往往含有的信息量非常大。其中部分甚至大部分信息对于行业具体应用而言意义不大。因此,如何提取有用信息,缩减视频的信息量就成为一个与实际应用紧密关联的重要问题。比如,在交通监控和安全监控视频中,使用视频对象分割技术将其中的主题信息(主要指运动对象信息)提取出来,才能更有效的对道路及生产环境安全进行监控。目前,视频对象分割技术就是最近几年发展起来的一种提取视频有效信息的重要基础性技术。该技术己广泛应用于交通流视频监控、工业自动化监控、安防、网络多媒体交互以及视频压缩编码等实际生产生活中。本文对视频对象分割技术及其应用进行了具体研究,并提出了一种基于高斯混合模型和小波理论的改进型的背景消减法。该方法有效削减了分割存在的噪声和空洞,提高了运动对象边缘的准确性。本文所做的主要工作如下:1.研究并实现了目前已有的多种分割算法。静态图像处理技术在视频分割中常用的工具及方法,如滤波方法、形态学算子及静态图像分割方法;视频对象分割技术,尤其是自动分割方法中的常用方法。2.研究并提出了一种改进型背景消减法。针对原有的基于高斯混合模型的背景消减方法存在噪声、空洞和边界分割不准确的不足。本文研究了一种由预分割与后处理组成的改进型背景消减法。其中,预分割基于高斯混合模型背景建模,并引入了图像的色彩模型和对比度模型,色彩模型主要用于提取每帧中大致的前景点,对比度模型主要用于确定前景和背景的边缘;后处理综合运用了中值滤波、小波理论中的金字塔分解和形态学开运算,以消除随机噪声和空洞。该方法有效地解决了原有背景消减法的不足,在削弱噪声和空洞的同时较准确地保留了对象边缘。3.研究了视频分割技术的实际应用。以现代交通中的道路监控为例,利用本文研究的算法,将视频中的车辆进行分割提取,然后使用轮廓逼近法,将分割出的对象加以标注,从而实现实时监控的辅助功能。最后,通过对本文算法与原有背景消减法及帧差法的的分割效果进行实验对比,验证了本文算法的优越性。更多还原

【Abstract】 In the information society, by providing the most extensive and comprehensive information elements, video content is more and more popular in the industry of modern traffic, network media and computer vision. However, the original video often contain too large amount of information. Some of them even for most of the information are not meaningful to industry specific applications. So, How to extract useful information, to reduce the amount of information from the video will be an important problem seriously related with practical applications. For example, in traffic control and safety monitoring video, only when the main information of the video (Mainly referring to the information of moving object) is extracted the work on monitoring road safety and the production environment will be well done.At present, video object segmentation technology developed in recent years is an important foundational technology in the area of extracting effective video information. The technology has been widely used in traffic streaming video monitoring, industrial automation control, security, network and multimedia interactive video compression coding. In this paper, video segmentation technology and its application were researched. And an improved background extinction method based on Gaussian mixture model and wavelet theory was proposed. The method effectively reduced the noise and holes in the segmentation and improves the accuracy of moving object.In this paper, the major work done is as follows:1. A variety of segmentation algorithm existed were researched and realized. The common tools and methods used in the video segmentation from static image processing technology, such as filtering method, morphological operators and static image segmentation method; video object segmentation technology, especially the common methods in the automatic segmentation methods.2. An improved background extinction was researched and proposed. The background extinction segmentation method based on Gaussian mixture model had the disadvantages of holes and inaccuracy in the segmentation edge. An improved background extinction method which contained pre-segmentation and post-processing was proposed. Pre-segmentation was based on Gaussian mixture model first, and then color model and contrast model was used in this method. Post-processing used several methods, such as median filtering, pyramid decomposition of wavelet theory, morphological open computing, to eliminate random noise and holes. This method effectively solved the disadvantages of the original background extinction method. When it eliminated holes and noises the edge was well maintained.3. Practical application of video segmentation was researched. As an example of road monitoring in the modern traffic, the method proposed in this paper was used first, the cars was segmented, and then contour approximation was used to labeled the object, so real-time tracking and auxiliary monitoring was realized. In the end, the method proposed in this paper, original background extinction method and frame differencing method were compared to prove the effectiveness of the former.更多还原