Learning action descriptors for recognition

摘要: This paper evaluates different Restricted Boltzmann Machines models in unsupervised, semi-supervised and supervised frameworks using information from human actions. After feeding these multilayer with low level features, we infer high-level discriminating features that highly improve the classification performance. approach eliminates difficult process of selecting good mid-level feature descriptors, changing selection extraction by a learning stage. Two main contributions are presented. First, new sequence-descriptor accumulated histograms optical flow (aHOF) is Second, comparative results experiments shown. The show RBM architectures provide very our task present properties for learning.