足球比赛视频中的目标检测与跟踪

【作者】 彭亮；

【导师】 娄震；

【作者基本信息】 南京理工大学 ， 计算机应用， 2006， 硕士

【摘要】 体育节目因为其特有的魅力，在现今的社会中广受大众的欢迎。因此，对体育比赛视频数据的分析研究，具有很高的实际意义和商业价值。本文以足球比赛视频为例，研究了比赛视频中感兴趣的目标——球员、足球和球场标志线的分割提取和预测跟踪。实际应用证明，这三类目标所包含的比赛的信息量最大。 目标分割提取时，本文针对目标的颜色、形状特点，分别采用不同的方法对球场线、球员和足球进行分割检测。对在球场线的分割检测方面，本文对比分析了顶帽变换和边缘检测算子在球场线分割上的效果。由于Hough变换具有对直线残缺和噪声的不敏感性，本文采用该变换方法获取球场线的坐标，并针对Hough变换不能得到直线端点的问题，讨论了一种直线端点的获取方法。同时本文将球场线的坐标在灰度图上进行拟合，以获得更为精确的球场线坐标参数。在球员和足球目标的分割提取方面，本文实现了一系列基于球员和足球的图像分割、候选区域提取以及识别分类的算法。针对图像分割时出现的球员区域中间断裂问题和球员球场线相互粘连问题，本文分析并提出了相应的解决方案。经过大量的实验证明，本文中的目标分割提取算法对噪声具有一定的鲁棒性，在不同的场地情况下，均能获得较为满意的效果。 在目标的预测跟踪时，对于球员和足球的跟踪，本文首先采用Kalman滤波对目标的运动轨迹进行预测估计，结合设置自适应波门来检测目标，实现对球员和足球的跟踪。针对在跟踪中出现的球员匹配和球员间相互遮挡的问题，本文研究并提出了可行的解决方案。对于球场线的预测跟踪，本文提出通过球场线的坐标参数来对球场线的运动趋势进行估计，在球场线的预测区域内进行直线提取操作。同时根据预测的坐标参数，限定Hough变换中两个参数的取值范围，以减小计算量。实验证明，利用本文的跟踪方法可以有效的对足球比赛视频中的目标进行跟踪，使得目标检测的计算量大大减少，其精度也有显著的提高。 本文提出的以足球比赛视频为例的体育视频分析处理技术，已经在实际工程中的虚拟广告生成和虚拟摄像机特效两个方面得到了应用。更多还原

【Abstract】 Nowadays, the programs of sports video are becoming more and more popular. Therefore, it possesses high actual meaning and commerce worthiness to analyse on the sports video. In this paper, soccer video is analyzed as an example. Detection and tracking of objects in the soccer video are studied. Here the objects we pay attention to include soccer players, ball and court lines. It is proved actually that these three kinds of objects contain a lot of information of the game.According to the difference in color and sharp, two different measures are adopted to detect the objects we focus on in this paper. About the detection for court lines, this paper discusses and compares the segment effects produced by the Top-Hat transform and the edge detectors. Insensitiveness to the noise, Hough transform is used to obtain coordinate parameters of lines. A method is discussed to get start and end points of lines. About the detection for players and ball, this paper advances a series of algorithms to segment, distill and recognize the areas of players and ball. The problems in segmented images that a clearance often exists in a player area and the conglutination between players and lines are discussed and an appropriate method is advanced to resolve it. Experimental results show that the detection algorithms advanced in this paper is robust to the noise. The results of objects detection are gratifying for various soccer video.This paper present an algorithm of tracking the objects in the soccer video based on Kalman filter. The tracking history is encoded into state parameter of the filter. The estimated state parameter will be used to reduce search area for model matching. The problems appearing in the tracking process about players matching and players sheltering are studied and methods to resolve them are present. This paper advances that tracking the court lines by estimating coordinate parameters of lines. And the parameters in Hough transform are restricted based on the estimated coordinate parameters to reduce the computation. Experimental results show that the tracking strategy presented in this paper is effective to track objects in soccer video.The process technologies for the sports video in this paper are already applied in actual projects.更多还原