A NOVEL APPROACH FOR ACTION RECOGNITION IN VIDEOS BASED ON SPATIOTEMPORAL CORNERS AND OPTICAL FLOW ESTIMATION

We provide a novel technique for locating the action in videos. In order to recognise that, we extract the STIP and PLK features. Their foundation is the identification of spatiotemporal corners. A location with a significant degree of change in picture intensity in all three directions is where spatial and temporal corners are found. This necessitates the placement of spatio-temporal corners at spatial corners that have large temporal gradient variation and invert motion in two successive frames. They may be distinguished from one another by local maxima in a corner ness function that has been calculated for all pixels at all spatial and temporal dimensions. For the estimate of optical flow, the Lucas-Kanade approach is a popular differential technique. It makes the assumption that the flow is largely constant in the immediate vicinity of the pixel under consideration and uses the least squares criteria to solve the fundamental optical flow equations for every pixel in that area.