Action Recognition based on Spatio Temporal SIFT detector
Action recognition in the realistic videos is a challenging problem in the computer vision. This paper presents a method of automatically recognizing the action performed by the human. The Spatio Temporal Scale Invariant Feature Transform (ST- SIFT) algorithm is made use of, for extraction of the keypoints in both spatial and temporal domain which is the extension of the 2D SIFT detector. The Spatio-Temporal Difference-of-Gaussian (STDoG) pyramid is initially built which is further used to find the maxima and the minima points that give the interest points. The keypoints are found in the xy,xt and yt planes where xy corresponds to the spatial plane, xt and yt planes correspond to the temporal domains. Experiment was conducted on a video containing a single action.
M Vinutha , V.S Veena Devi