Systems and methods for motion estimation for coding a video sequence

A system for motion estimation for coding a video sequence having a plurality of frames, the system comprising:

• a memory for storing data, metadata and one or more programming modules;
• and
• a processor configured for executing the one or more programming modules stored in a memory that upon execution executed the steps of:
• extracting one or more macro block from each of the plurality of frames;
• determining for each of the one or more macro blocks a corresponding motion vector by dynamically performing motion estimation;
• determining for each of the plurality of frames a candidate macro block by:
• calculating a motion classifier weight factor for each of the one or more macro blocks of a current frame, the motion classifier weight factor is weighted average of a block movement factor for each of the one or more macro blocks of a previous frame corresponding to a current frame and a temporal redundancy factor for each of the one or more macro blocks of the previous frame;
• calculating a link bandwidth weight factor upon receiving an Instantaneous Decoder Refresh frame;
• calculating a probability factor, the probability factor being the probability of selecting search pattern for the current frame same as the search pattern of the previous frame;
• determining a search method coefficient for the current frame, wherein the search method coefficient is a function of the motion classifier weight factor, transform of the link bandwidth weight factor, and the probability factor;
• and
• dynamically selecting a block matching method for searching a candidate macro block based on the search method coefficient, wherein
• if the motion classifier weight factor indicates very high motion and the transform of the link bandwidth weight factor is less than a video generation rate then a Non uniform hierarchical search is performed, the Non uniform hierarchical search includes a combination of integer pixel search and half pixel or quarter pixel search with different types of search patterns, in case the transform of the link bandwidth weight factor is greater than the video generation rate and the probability factor is greater than a threshold then the search pattern of the previous frame is selected for the present frame, else a Variable hexagon based search is performed;
• if the motion classifier weight factor indicates high motion and the transform of the link bandwidth weight factor is less than the video generation rate then the Variable hexagon based search is performed, in case the transform of the link bandwidth weight factor is greater than the video generation rate and the probability factor is greater than a threshold then the search pattern of the previous frame is selected for the present frame, else a Fixed 3-step small hexagon based search is performed, the Fixed 3-step small hexagon based search is completed in 3 steps by searching at the points of hexagon centered with the corresponding current macro block in the previous frame;
• if the motion classifier weight factor indicates medium motion and the transform of the link bandwidth weight factor is less than the video generation rate then the Fixed 3-step small hexagon based search is performed, in case the transform of the link bandwidth weight factor is greater than the video generation rate and the probability factor is greater than a threshold then the search pattern of the previous frame is selected for the present frame, else a Variable diamond based search is performed;
• and
• if the motion classifier weight factor indicates low motion and the transform of the link bandwidth weight factor is less than the video generation rate then the Variable diamond based search is performed, in case the transform of the link bandwidth weight factor is greater than the video generation rate and the probability factor is greater than a threshold then the search pattern of the previous frame is selected for the present frame, else a Fixed 3-step small diamond based search is performed, the Fixed 3-step small diamond based search is completed in 3 steps by searching at the points of diamond centered with the corresponding current macro block in the previous frame;
• and
• encoding the video sequence based on the candidate macro block.

A non-transient computer-readable medium, including instructions stored thereon which, when read and executed by one or more computers, causes the one or more computers to perform the method comprising the steps of:

• extracting one or more macro block from each of the plurality of frames;
• determining for each of the one or more macro blocks a corresponding motion vector by dynamically performing motion estimation;
• determining for each of the plurality of frames a candidate macro block by:
• calculating a motion classifier weight factor for each of the one or more macro blocks of a current frame, the motion classifier weight factor is weighted average of a block movement factor for each of the one or more macro blocks of a previous frame corresponding to a current frame and a temporal redundancy factor for each of the one or more macro blocks of the previous frame;
• calculating a link bandwidth weight factor upon receiving an Instantaneous Decoder Refresh frame;
• calculating a probability factor, the probability factor being the probability of selecting search pattern for the current frame same as the search pattern of the previous frame;
• determining a search method coefficient for the current frame, wherein the search method coefficient is a function of the motion classifier weight factor, transform of the link bandwidth weight factor, and the probability factor;
• and
• dynamically selecting a block matching method for searching a candidate macro block based on the search method coefficient, wherein
• if the motion classifier weight factor indicates very high motion and the transform of the link bandwidth weight factor is less than a video generation rate then a Non uniform hierarchical search is performed, the Non uniform hierarchical search includes a combination of integer pixel search and half pixel or quarter pixel search with different types of search patterns, in case the transform of the link bandwidth weight factor is greater than the video generation rate and the probability factor is greater than a threshold then the search pattern of the previous frame is selected for the present frame, else a Variable hexagon based search is performed;
• if the motion classifier weight factor indicates high motion and the transform of the link bandwidth weight factor is less than the video generation rate then the Variable hexagon based search is performed, in case the transform of the link bandwidth weight factor is greater than the video generation rate and the probability factor is greater than a threshold then the search pattern of the previous frame is selected for the present frame, else a Fixed 3-step small hexagon based search is performed, the Fixed 3-step small hexagon based search is completed in 3 steps by searching at the points of hexagon centered with the corresponding current macro block in the previous frame;
• if the motion classifier weight factor indicates medium motion and the transform of the link bandwidth weight factor is less than the video generation rate then the Fixed 3-step small hexagon based search is performed, in case the transform of the link bandwidth weight factor is greater than the video generation rate and the probability factor is greater than a threshold then the search pattern of the previous frame is selected for the present frame, else a Variable diamond based search is performed;
• and
• if the motion classifier weight factor indicates low motion and the transform of the link bandwidth weight factor is less than the video generation rate then the Variable diamond based search is performed, in case the transform of the link bandwidth weight factor is greater than the video generation rate and the probability factor is greater than a threshold then the search pattern of the previous frame is selected for the present frame, else a Fixed 3-step small diamond based search is performed, the Fixed 3-step small diamond based search is completed in 3 steps by searching at the points of diamond centered with the corresponding current macro block in the previous frame;
• encoding the video sequence based on the candidate macro block.