PedVed: Pseudo Euclidian distances for video events detection

Abstract

This paper provides a new method that generates automatically pseudo Euclidian distances (PED) from the trigonometrically treatments of motion history blobs (MHB) obtained from motion history images (MHI) to extract efficient image features, which are pertinent to video events detection (VED). Given a point with its direction of motion where the point coincides the center of a circle. How far the point can virtually travel inside the circle with that direction? That virtual distance is called pseudo Euclidian distance. PED, would be potentially used in wide variety of computer vision applications, remains the main contribution of this paper. To show the interest of the usage of PED, we have proposed a PED based methodology for VED and the detection results of some events at TRECVID'08 in real videos have been demonstrated. © 2009 Springer-Verlag.