Dynamic motion vector analysis method

#The present invention relates to a method to optimize bandwidth requirements and on-chip memory usage by Dynamic Motion Vector Analysis for motion compensation based video decoder systems and apparatus thereof.

#In motion-compensation methods, previously decoded frame data (called reference data) is required while decoding the current frame. Since on-chip memory is a precious resource, even for moderate size frames it is required to store reference (previously decoded) frames in the external memory. When required by motion compensation methods this data needs to be fetched from external memory and this leads to additional bandwidth requirement on the interface. The data to be referenced by adjacent regions can have some overlap and if by having some extra on-chip memory this overlap can be exploited then it reduces some bandwidth at the expense of some on-chip memory. For higher frame sizes the bandwidth overhead can be prohibitive and it necessitates having a significant on-chip memory for the reference frames. A method, called Dynamic MV-Analysis, which defines a reference-data fetch strategy is proposed. This method dynamically achieves a balance between external memory interface bandwidth requirement and on-chip memory requirement thus leading to significant savings by exploiting data overlaps and reducing redundant loads.

#While referencing data from pervious frames multiple scenarios can be there. A few of these are: Current data blocks referencing overlapping data from previous frames as is the case in low motion frames. Current data blocks referencing non-overlapping (reference regions far apart) regions. This will be the case in high motion frames with large motion vectors. Current data blocks referencing regions from multiple reference frames.

#The invention will be described in greater detail below in connection with accompanying drawing figures, in which:

#FIG. 1 shows the coordinates of the reference region to be fetched, {circumflex over (R)}lri, corresponding to each Blri in a given region.

#FIG. 2 shows flow chart for MV-analysis method.

#The primary embodiment of the present invention is a method to optimize bandwidth requirements and on-chip memory usage by Dynamic Motion Vector Analysis for motion compensation based video decoder systems, said method comprising steps of; providing region for which data is to be fetched; identifying reference region from the regions stored in external memory using motion vectors, consisting pre-processing motion vectors to determine reference data fetches from multiple reference frames; identifying smallest overlap region for each of the reference data blocks; comparing the overlap region with predetermined threshold; and fetching the regions having value less than the threshold and dividing the regions having value more than the threshold into sub-regions and repeating the steps of identifications, comparing, and fetching till each region cannot be further sub-divided.

#In yet another embodiment of the present invention, the size and number of the regions depends upon a set of predetermined thresholds.

#In still another embodiment of the present invention, the method which provides for means to trade-off between external interface bandwidth requirement and the number of data fetches for a given amount of on-chip memory in a system.