Output details
11 - Computer Science and Informatics
University of Dundee
Action categorization with modified hidden conditional random field
<23> This paper advanced the hidden conditional random field (HCRF) for action recognition. Specifically, effective silhouette-based action features are extracted using motion moments and spectrum of chain code. A modified HCRF (mHCRF) is then formulated to have a global optimum in the modelling of the temporal action dependencies after the pathing stage. Rigorous experimental results on action categorization using this model are compared favourably against several existing model- based methods including GMM, SVM, Logistic Regression, HMM, CRF and HCRF. This work is supported by INSIGHT (GR/S63687/01) project funded by EPSRC and MOD DSTL