《特征标志检测与场景识别技术在体育视频中的应用研究》由会员分享，可在线阅读，更多相关《特征标志检测与场景识别技术在体育视频中的应用研究（58页珍藏版）》请在金锄头文库上搜索。
1、杭州电子科技大学 硕士学位论文 特征标志检测与场景识别技术在体育视频中的应用研究 姓名：王磊 申请学位级别：硕士 专业：计算机软件与理论 指导教师：陈临强 20091201 杭州电子科技大学硕士学位论文 I 摘 要 体育比赛的主办方为了充分发掘广告带来的商业利润，总是希望尽可能地在比赛场地周 围放置更多的广告牌。但是在实际的比赛现场，出于保护运动员和保证观众观看质量等方面 的考虑，比赛场地周围的很多区域是禁止摆放广告牌的。例如足球比赛中的底线球门附近， 网球比赛的底线和边线附近等。这时，转播方可以利用视频融合技术，在转播体育视频的同 时对其进行分析，将预先准备好的广告图片或动画作几何形状上的变
2、换后，融合到视频图像 中指定的场地位置上，也可实时在虚拟场景下渲染虚拟物体，以形成类似于现场广告的虚拟 广告。 特征标志检测和场景识别技术是体育视频融合系统的核心技术，它分析体育视频中场地 的特征标志信息，并利用这些信息建立与真实场地之间的透视变换关系，为虚拟广告的融合 提供变换依据。 国外已经有不少公司对体育视频融合技术展开了研究，并有一些产品投入市场，在国内 也有部分研究机构对该技术展开了研究，已经有一些初步成果。由于体育视频丰富多彩，比 赛场地千差万别，同时存在天气环境变化和相机微运动等复杂不确定因素，因此在特征标志 检测和场景识别技术中有若干问题需要深入研究。 本文的研究内容主要包括以
3、下三点： 1、视频帧主区域的提取。由于场景中包含大量观众区域等干扰信息，去除干扰信息能明 显提高图像检测精度和检测速度，本文分析了比赛场地色彩分布特点，采用了基于直方图统 计的球场主区域自动提取方法，同时细节方面还包括噪点滤波和空洞弥补两方面。 2、场地特征标志点的检测。特征点的来源主要以特征线的交点为主，针对微运动相机下 特征点运行的实际特点， 场地特征点的检测分成首次初始化全局检测和运行相机下局部检测， 其中运动相机下的特征点检测对速度和精度都有很高的要求，在实现上主要采用运动点 Kalman 预估和预测点局部检测相结合的方法。 3、相机微运动下透视变换关系稳定性分析。相机微运动、低分辨率
4、、误差扩大等多重因 素造成虚拟信息的抖动现象（融合不稳定） ，该现象不是单纯依靠改进特征信息检测算法而提 高检测精度所能解决的，而根据透视变换原理，分析特征信息前后帧关联关系，从而排除部 分相对误差大的特征信息，依靠这种方法能从本质上提高视频融合的稳定性。 关键词：视频融合，运动相机，场景识别，虚拟广告，特征标志检测 杭州电子科技大学硕士学位论文 II ABSTRACT In order to excavate the commercial ads profits fully, sports competitions organizers always hope that lays aside
5、 more billboards as far as possible around the competition range. However, in the actual game scene, for the protection of athletes and the quality spectators watched, many regions forbid to place the billboard around competition range. For example, nearby agent goal in football competition, nearby
6、agent and sideline in tennis match and so on. At this time, broadcast company can use the video fusion technology, analysis the sports video while broadcast sports video, and integrate the pre-prepared images or animated ads into the video image, and can also render the virtual objects to the virtua
7、l scene in real-time, so as to generate the virtual advertisement which is similar to the real advertisement. The characteristic symbol detection and the scene recognition technology is the core technology of sport-video frequency fusion system, it analyzes the characteristic symbol information in t
8、he sport-video, and establishes the perspective transformation which relates to the real location and provides the transformation basis for the hypothesized advertisements fusion by using the information. Many overseas companies have a lot of research about sport-video fusion technology, and some pr
9、oducts have been put into the market. There is also some development facility launch the research to this technology, and already have some initial fruits. There are many issues need to research in-depth in characteristic markers and scene recognition technology, because sport-video is colorful、The
10、competition range is infinitely varied and there is also uncertainties factors about micro-environmental changes and cameras movements. The research content of this paper covers the following three points. First, the extraction main region in video frequency frame. As the scene contains a large numb
11、er of disturbance information audiences such as information of audiences region, remove disturbance information can significantly improve the image detection accuracy and detection speed. This paper analyzes the characteristic of competition range, and design a automatic method to extract the main r
12、egion based on color histogram statistics which also includes noise filtering and holes atonement. Second, characteristic marked point examination in location. The lines intersection points are the primary origin of characteristic points, in view of the actual characteristic of characteristic point
13、movement under micro-movement camera, the detection of characteristic points divides into Characteristic point overall situation examination in the first time and characteristic point partial 杭州电子科技大学硕士学位论文 III examination under micro-movement camera. Characteristic point partial examination under m
14、icro-movement camera need very high request of the speed and the precision, so this paper use the method that complex with partial examination relate to characteristic points and Kalman estimating relate to motor points. Third, stability analyses of transform relations under micro-movement camera. T
15、he multiple factors include micro-movement camera、 Low resolution、 Erroneous expansion create the vibration phenomenon (do not fuses stably)in the hypothesized advertisement, This phenomenon can not be solved only by the improvement of characteristic information detection algorithm increases the det
16、ection precision, except the method that analysis interrelatedness of characteristic information in around frames, exclusive characteristic information points which have lager relative error, and it is able to enhance the stability essentially of the advertisement fusion depends upon this method. Keywords: video fusion, motion camera, scene recognition, virtual advertising, feature marked 杭州电子科技大学杭州电子科技大学 学位论文原创性声明和使用授权说明学位论文
