Change detection in digital video signals

Abstract

We present a method for scene change detection based on projections onto the vertical, horizontal and diagonal axes. At first, vertical projection of each two consecutive interframe differences are calculated. Then based on the distance measure between them, together with proportionality and sign tests, fade in/out, dissolve, wipe and cut can be classified. We also propose a method for small moving object detection in video sequences based on adaptive prediction and higher order statistical tests. We first eliminate camera movement using subpixel accurate motion estimation. Then adaptive prediction is applied on the image obtained from motion compensation and an error image is obtained. Higher order statistical test is then applied on the residual image to detect small moving objects whose size may consist of only a few pixels.