Analysis of camera operations in MPEG compressed domain based on generalized hough transform

Abstract

In this paper, we propose a simple and efficient method to estimate the camera operations using MVs(Motion Vectors), which is extracted directly from MPEG-2 video stream without complete decoding. In the method, MVs in the compressed video stream are converted into approximated OF(Optical Flow). And the approximated OF is used to estimate the camera operations including pan, tilt and zoom based on the generalized Hough transform technique. The method provided better results than the conventional least square method for video streams of basketball and soccer games including captions. And these camera operations can be used in constructing a key frame as a mosaics image or making features for content-based video retrievals.