Parallel Motion Estimation Region Merge Candidate Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The current High Efficiency Video Coding (HEVC) merge mode design is sequential, which hinders parallel motion estimation, leading to coding efficiency loss and increased computational complexity due to the lack of availability of neighboring motion data for inclusion in a merge candidate list during parallel processing.
Innovation Solution
A video coding device and method that selects a subset of motion vector candidates based on priority levels to generate a merge candidate list, improving coding efficiency and reducing computational complexity by including only relevant motion data in the list.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If sequential merge mode design is used, then motion data availability is improved, but processing speed and productivity deteriorate
Solution Approach 1:
The motion estimation region is divided into multiple parallel motion estimation regions (MERs), each processed independently by different processing elements. This segmentation allows simultaneous processing of multiple prediction units without requiring sequential data availability, thus improving processing speed while maintaining adequate motion data availability within each segment.
Solution Approach 2:
Motion vectors for boundary prediction units are computed in advance before parallel processing begins. These pre-computed motion vectors are stored and made available as merge candidates for subsequent parallel processing, eliminating the need to wait for sequential computation and enabling faster parallel execution.
2Manufacturing precision
If all motion vector candidates are included in merge candidate list, then coding precision is improved, but device complexity increases
Solution Approach 1:
The patent introduces a priority level parameter for motion vector candidates, transforming the selection process from including all candidates to selecting based on priority thresholds. This parameter change enables controlled reduction of candidate list size while maintaining coding precision by prioritizing the most relevant motion vectors.
Solution Approach 2:
The patent extracts and prioritizes only the most relevant motion vector candidates based on their priority levels, removing less important candidates from the merge candidate list. This extraction process reduces the number of candidates that need to be processed and stored, thereby reducing device complexity while preserving coding precision through selective inclusion of high-priority candidates.
3Productivity
If parallel motion estimation is implemented, then processing speed is improved, but motion data availability deteriorates
Solution Approach 1:
By dividing the processing into separate parallel MERs, each processing element has access to motion data within its segment without needing to wait for other segments. This segmentation maintains motion data availability within each parallel unit while enabling simultaneous processing across multiple units, improving overall processing speed.
Solution Approach 2:
Motion vectors required for parallel processing are pre-computed and stored before the parallel estimation begins. This preliminary action ensures that all necessary motion data is available at the start of parallel processing, eliminating data availability bottlenecks while maintaining high processing speed throughout the parallel execution.
Data Source
AI summary
A video coding device configured according to some aspects of this disclosure includes a memory configured to store a plurality of motion vector candidates. Each motion vector candidate can corresponding to at least one of a plurality of prediction units (PUs) partitioned in a parallel motion estimation region (MER). The video coding device also includes a processor in communication with the memory. The processor is configured to select a subset of the plurality of motion vector candidates to include in a merge candidate list. The selection can be based on a priority level of each motion vector candidate. The processor can be further configured to generate the merge candidate list to include the selected motion vector candidates.