A Novel Model for Human Activity Representation and Classification Based on Trajectory
Wei Xing Zhu, Wei Guang Xu, Guo Lin Hou, Jun You · Advanced materials research · 2014
In this paper, a novel model, HDP-HMM-SCFG is proposed for representing and classifying activities based on motion trajectories. In the model, activities are represented by stochastic grammar using trajectory, where trajectory segments are considered as observations emitted by the grammar terminals attached with HMMs. Then, by replacing the Euclidian distance in the kernel function of Gaussian radial radix with EMD-DTW, which is proposed to measure the distance between two trajectories by integrating the pros. of both EMD and DTW, multi-class SVM classifier is constructed. Experiments on ASL dataset are carried on to validate our approach.