An efficient approach for multi-view human action recognition based on bag-of-key-poses
HBU'12 Proceedings of the Third international conference on Human Behavior Understanding
| Hi-index | 0.00 |
A novel algorithm for view-invariant human action recognition is presented. This approach is based on Two-Dimensional Principal Component Analysis (2DPCA) applied directly on the Motion Energy Image (MEI) or the Motion History Image (MHI) in both the spatial domain and the transform domain. This method reduces the computational complexity by a factor of at least 66, achieving the highest recognition accuracy per camera, while maintaining minimum storage requirements, compared with the most recent reports in the field. Experimental results performed on the Weizmann action and the INIRIA IXMAS datasets confirm the excellent properties of the proposed algorithm, showing its robustness and ability to work with small number of training sequences. The dramatic reduction in computational complexity promotes the use in real time applications.