A Real-time Billboard Trademark Recognition Algorithm in Sports Video

Recently, in the sports video analysis domain, applications like TV commercials can be developed by detecting and recognizing the trademark on the billboard. In this paper, we propose a method for real-time trademark recognition in different sports video such as soccer, basketball, and Formula 1. There are two stages in this algorithm. Fuzzy decision tree based method is used to detect billboard frame in the first stage, while in the second stage, color and regional SIFT (scale-invariant feature transform) features are combined to describe the appearance of trademarks, and latent semantic analysis (LSA) based SIFT vocabulary matching is performed to recognize the given template trademark. The preliminary experiments demonstrate the effectiveness and efficiency of our algorithm.