Players Localizatio , Highlighti G a D Team Classificatio I Soccer Game Video

In this paper, a framework of player segmentation, tracking, highlighting and team classification in soccer game video is proposed. Main contributions consist of: 1) playfield modeling combining the Gaussian model with the dominant color extraction to segment the players/referees; 2) mean shift-based tracking with soft constraints from the foreground map to alleviate the shortcomings in handling the fast moving players; 3) discriminant similarity metric de-weighted by the surrounding background distribution applied for scale adaptation; 4) conservative update of the target model by aligning with its initial; 5) bi-histogram matching employed for team classification. Experimental results are given to demonstrate the proposed approach’s performance.