The availability of low-cost range sensors and the development of relativelyrobust algorithms for the extraction of skeleton joint locations have inspiredmany researchers to develop human activity recognition methods using the 3-Ddata. In this paper, an effective method for the recognition of humanactivities from the normalized joint trajectories is proposed. We represent theactions as multidimensional signals and introduce a novel method for generatingaction templates by averaging the samples in a "dynamic time" sense. Then inorder to deal with the variations in the speed and style of performing actions,we warp the samples to the action templates by an efficient algorithm andemploy wavelet filters to extract meaningful spatiotemporal features. Theproposed method is also capable of modeling the human-object interactions, byperforming the template generation and temporal warping procedure via the jointand object trajectories simultaneously. The experimental evaluation on severalchallenging datasets demonstrates the effectiveness of our method compared tothe state-of-the-arts.
展开▼
