Figure 5.
The proposed hierarchical activity recognition framework with feature selection integrated. In the phase of training, it selects the optimal feature subset and builds a classifier for each of the non-leaf nodes. In the phase of prediction, it adopts a top-down fashion to give the predicted label.