Human action recognition based on two-view optical flow in the transformed domain

In this paper a human action recognition algorithm based on two-view of optical flow in multiple layers per camera employing the transform domain and 2DPCA is presented. This method explores more distinctive features between actions. It is not sensitive to translation, alignment and noise. In addition the use of 2DPCA maintains the spatial relation between pixels and increases the recognition accuracy. Experimental results performed on the Weizmann and IXMAS datasets confirm these excellent properties compared to recent reported methods.