This is the process of analyzing previous frames and next frames to identify blocks that have not changed or have moved location, during the encoding of a video stream. It’s important that error is kept to a minimum in this process as it could alter the playback badly.
Depending on the video type, users can use Quarter Pixel Precision (Qpel) or Half Pixel Precision for Motion Estimation. For example, Quarter Pixel Precision (Qpel) for Motion Estimation is used to more accurately Track movement across the video Frame. Although entire Pixels must be stored, when calculating locations for a moving object the motion from one frame to the next may be less than a single pixel. MPEG-2 uses Half Pixel Precision.
When motion estimation is performed by an MPEG-2 encoder it groups pixels into 16x16 macroblocks. MPEG-4 AVC encoders can divide these macroblocks into partitions as small as 4x4, and even of variable size within the same Macroblock. For example, a single macroblock could have 2 4x4 Partitions and one that's 8x4. They must be divisible by 4 in both dimensions.
Partitions allow for more accuracy in motion estimation because areas with high motion can be isolated from those with less movement.