Global motion estimation algorithm for video segmentation

This work presents a new global motion estimation algorithm for MPEG compressed video sequences. It makes use of a feature extraction technique based on the Generalized Hough Transform, which is able to provide rotation, scale and displacement parameters when comparing two different frames from a video sequence. Thus, pan, tilt, swing (rotation along z-axis) and zoom effects will be studied using the proposed algorithm. DC coefficients of the DCT transform are extracted from the MPEG stream and used to create DC images, which are the starting point for the global motion estimation algorithm. Further application of the feature extraction technique to DC images will allow to perform motion estimation, reducing processing time as the decompression process is avoided. Pseudocode and details about the implementation of the algorithm as well as statistics to illustrate the efficiency of the algorithm are provided.