The DMPAC SDE aggregates local cost in a custom semi global method. It follows the following steps:
- Input image intensity data is first transformed into Census domain by Census
Transform .
Census transformation is performed over an 9x9 pixel neighborhood
window.
- The local matching cost between two image pixel positions are measured by the Hamming Distance of the corresponding census signature.
- Local costs are then aggregated using a custom semi global matching (SGM) method.
- Similar to standard SGM methods, penalty terms (P1, P2) for disparity changes in the local neighborhood is used.
- The penalty terms P1, P2 are used in a non-adaptive way, that is, they remain constant for the entire frame processing.