自适应高斯混合模型球场检测算法及其在体育视频分析中的应用

球场检测在体育视频内容分析中有着重要作用.为了克服由于不同光照、不同相机、不同拍摄角度造成球场颜色的非均一性问题，提出了一种基于自适应高斯混合模型(adaptive Gaussian mixture model, GMM)的球场检测算法.该算法首先从视频中任意抽取一些图像，并自动分析这些图像的主要颜色，从中找到主颜色的近似分布，然后，利用GMM拟合主要颜色分布.为提高模型的适应能力，在球场检测过程中，利用当前GMM球场检测结果和增量期望最大(incremental expectation maximum, IEM)算法不断更新模型参数，从而得到更加准确的参数估计，并用于后续图像中球场和非球场像素进行分类.最后，根据球场区域在图像中的分布，对足球比赛场景进行分类.实验证明，提出的算法具有良好的性能.

Playfield Detection Using Adaptive GMM and Its Application in Sports Video Analysis

Playfield detection is a key step in sports video content analysis, while the playfield presents different colors in broadcast video as a result of illumination variation caused by different cameras and different shooting angles, etc. To address the problem, adaptive GMM is exploited to detect playfield. The proposed algorithm first selects samples in a video clip randomly and extracts the dominant color in the histogram automatically; then the dominant color is modeled by GMM approximately. To adapt the model to the playfield color variation, the model's parameters are updated by incremental EM algorithm in the process of playfield detection. Based on the playfield's detection result, a simple and effective algorithm is proposed to classify the scene in soccer video according to the playfield distribution region in a frame. Experimental results show that the proposed algorithms are effective.