Noisy motion vectors removal for reliable camera parameters estimation in MPEG coded videos

We face the problem of the estimation of camera motion parameters from MPEG compressed streams. The most common approach relies on the use of the motion vectors defined in the P and B coded frames of the MPEG streams. In this way, the computational efficiency of the whole process is improved. However, in some specific situations the motion vectors do not represent the real motion of the intercoded macroblock; thus, it is important to detect and remove such noisy macroblocks (outliers) in order to obtain reliable camera motion estimation. Hence, we present a simple method for outliers removal. Some preliminary experiments are reported that demonstrate the effectiveness of the proposed technique.