The block matching process is based on
motion predictors and two step coarse-to-fine refinement. Where:
- Predictor search stage evaluates configured optical flow predictors and
selects the best perdictor that minimizes the block matching cost.
- DMPAC defines a list 7 motion predcitors that are supported and can be
configured at pyramid level. These are reffered to as TopLeft, Top,
TopRight, Left, PyramidalLeft, PyramidalCurrent and Temporal predictors.
- DMPAC allows use of upto 5 motion predictors for a pixel and 2 of
these can be configured at each pyramid level and other three are
fixed.
- Upper pyramid levels (L5-L1) can use any 2 of Left, PyramidalLeft
and PyramidalCurrent predictors.
- Base pyramid level (L0) can use any 2 of Left, PyramidalLeft
PyramidalCurrent and Temporal predictors.
- Two step coarse to fine step search method refines optical flow estimates
around winner predictor. A configurable motion smoothness factor value
allows the block matching process to control the smoothness of the flow
field.