Output details

Boosted key-frame selection and correlated pyramidal motion-feature representation for human action recognition

We proposed a novel method for action recognition via boosted key-frame selection and correlated pyramidal motion features. It's the first time that AdaBoost was used to select keyframes from video sequences for action representation in a discriminative way, which dramatically reduces the complexity for action representation and enhances the discriminative power of recognition systems. The proposed new descriptor is more informative and robust than previous features. The whole framework was evaluated on three challenging datasets and outperforms state-of-the-art methods. The algorithm has the potential to tackle real-world action recognition problems with applications in video surveillance, video search/retrieval, human-machine interaction, etc.