Hardware for Motion Estimation

Abstract

Typically, two consecutive frames of a video sequence are very similar. This observation is the basis of motion-compensated coding, where a frame is coded based on its difference from another frame. In practice, frames are divided into blocks. For a block in the current frame (called the reference or source block), motion estimation is the process of finding a block in another frame that best matches (according to a given criterion) its characteristics. A motion vector identifies the position of the best block relative to the reference block. The search for the best matching block is done in a rectangular area (called the search window) around the relative position of the reference block. Figure 11.1 shows a search window that extends p pixels to the left and top and p — 1 pixels to the right and bottom of an M × N reference block. This is usually referred to as a [ —p, p — 1] search window. In most video applications, the reference block is 1616 pixels, and the search window is 31 × 31 pixels (p = 8).