Camera motion estimation using feature points in MPEG compressed domain

Abstract

A computationally efficient camera motion estimation algorithm based on feature points derived from compressed domain is presented. The feature point selection algorithm employs the texture/edge information contained in the DCT (discrete cosine transform) coefficients and is therefore directly applicable to DCT-based compressed video (like MPEG-1/-2/-4, H.261, H.263, DV or the MJPEG format). This paper describes: (1) the extraction of the feature-points in the compressed domain, (2) the application of computationally efficient motion estimation on these feature points by taking advantage of existing motion vectors in the MPEG compressed domain, as well as (3) an application for camera motion estimation.