The action recognition algorithm developed so far, uses the Moment based approach to recognize actions performed by a person in a video. The moments extracted gives us a global characteristic of the human body silhouette present in a video frame and therefore, recognizes only global actions. These moments computed from a sequence of video frames are analyzed using PCA and suitable basis vectors are obtained enabling us to project the test action onto this subspace for action classification.

Video Demonstrations

Action Recognition

VIPS – Vision Lab Project Demonstrations