Discriminative Action Recognition Using Supervised Latent Topic Model

Abstract

We present a discriminative learning method for human action recognition from video sequences. Our model combines a bag-of-words component with supervised latent topic models. The supervised latent Dirichlet allocation (sLDA) topic model, which employs discriminative learning using labeled data under a generative framework, is introduced to discover the latent topic structure which is most relevant to action categorization. We test our algorithm on two challenging datasets. Experimental results demonstrate the effectiveness of our algorithm.