Human action recognition with spatial-temporal feature filter

Spatial-temporal interest points (STIP) is the main method in human action recognition. However, there are some STIPs with low discriminative power for action recognition and weak ability for classification. To resolve the problem and achieve high recognition accuracy, this paper proposes a variance filter method for STIP selection. Finally, various experiments are conducted to prove that variance filter for STIP selection for action classification is superior to existing methods based on STIPs.