A statistical model based on spatio-temporal features for action recognition

Local spatio-temporal features have recently become a popular video representation for action recognition. In this paper, we propose a statistical model based on sparse representation of space-time features. The Harris3D detector, which extends the Harris detector for images to image sequences, is used as a feature detector, and histograms of gradient orientations (HOG) is used as a feature descriptor. The statistical distribution of the local spatio-temporal features for each action category is obtained using the independent component analysis (ICA). Finally, we test our model on public action database KTH, and the recognition results demonstrate the effectiveness of our model.