PLAYFIELD SEGMENTATION FOR BASEBALL VIDEOS USING ADAPTIVE GMMS
Chung‐Ming Kuo, Mao‐Hsiung Hung, Chih-Shan Liu, Yukon Chang, Chaur‐Heh Hsieh · 2010
Playfield is one of main parts appearing in typical scenes of sports videos. The segmentation of playfield is essential because it usually offers important cue of higherlevel content in sport videos. The colors in playfield may change significantly due to variation of illumination, which thus results in a large amount of segmentation errors for color-based segmentation schemes. In this paper, we present a new playfield segmentation method for baseball videos based on an adaptive Gaussian Mixture Model (GMM). The Expectation Maximization (EM) algorithm is used to train the model parameters. An adaptive GMM model is constructed by a novel training sample selection, which automatically selects appropriate samples from input video for the parameter estimation of EM process. The simulation results indicate that the new method achieves very low error rates of segmentation in an efficient manner.