Compressed domain action classification using HMM

The paper proposes three techniques for person independent action classification in compressed MPEG video. The features used are based on motion vectors, obtained by partial decoding of the MPEG video. The features proposed are projected 1D, 2D polar and 2D Cartesian. The feature vectors are fed to a hidden Markov model (HMM) for classification of actions. In total, seven actions were trained with distinct HMM for classification. Recognition results of more than 90% have been achieved. This work is significant in the context of the emerging MPEG-7 standard for video indexing and retrieval