ICA mixture hidden conditional random field model for sports event classification

In this paper, a new hidden conditional random field (HCRF) model with independent component analysis (ICA) mixture feature functions is developed for sports event classification. Unlike Hidden Markov Model (HMM), HCRF is a discriminative model without conditional independence assumption of observations, which is more suitable for video content analysis. According to the non-Gaussian property of sports event features, a new feature function using the likelihood of ICA mixture component is proposed to further enhance the HCRF model. The discriminant power of HCRF and representation power of ICA mixture for non-Gaussian distribution are combined. The new model is applied to challenging bowling and golf event classification. The simulation results prove our analysis that the new ICA mixture HCRF outperforms the existing mixture HMM models in term of classification accuracy.