Parallel Motion Estimation Regions in Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding standards, such as HEVC, face challenges in achieving efficient parallel motion estimation due to inter-dependencies in spatial motion data positions, leading to quality loss and increased processing time, especially when performing merge and skip modes.
Innovation Solution
The approach involves dividing the largest coding unit into non-overlapping parallel motion estimation regions, allowing for parallel motion estimation within each region while considering spatial motion data positions as unavailable for merging candidate list construction, and using alternative positions when necessary, to decouple merging candidate list construction and motion estimation from regular motion estimation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If spatial motion data positions are used for merging candidate list construction in parallel motion estimation, then motion estimation accuracy is improved, but processing time increases and quality loss occurs due to inter-dependencies
Solution Approach 1:
The LCU is divided into multiple non-overlapping parallel motion estimation regions (PMERs), allowing motion estimation to be performed independently in each region. This segmentation eliminates inter-dependencies between spatial motion data positions, enabling parallel processing while maintaining motion estimation accuracy within each region.
Solution Approach 2:
The patent introduces a new dimension of parallelization by creating multiple independent PMERs within the LCU structure. This allows motion estimation to proceed simultaneously across different regions rather than sequentially processing spatial neighbors, thus reducing processing time without sacrificing accuracy.
2Manufacturing precision
If regular motion estimation is performed sequentially considering all spatial motion data positions, then coding quality is maintained, but throughput requirements are not met
Solution Approach 1:
By segmenting the LCU into multiple PMERs, the patent enables parallel processing of motion estimation while maintaining coding quality within each region. The segmentation allows independent processing streams that can be executed simultaneously, thereby increasing throughput without compromising the quality of motion compensation.
Solution Approach 2:
The patent applies local quality by ensuring that each PMER maintains adequate motion estimation quality through independent candidate list construction, while the overall LCU achieves high throughput through parallel processing of multiple regions. Each region is optimized for both quality and processing efficiency.
3Productivity
If spatial motion data positions are made unavailable for merging candidates in PMER, then parallel processing is enabled, but candidate list construction complexity increases
Solution Approach 1:
The segmentation of LCU into PMERs with unavailable spatial motion data positions actually simplifies candidate list construction by eliminating the need to check for inter-dependencies between regions. Each PMER independently constructs its candidate list without complex coordination, reducing overall system complexity while enabling parallel processing.
Data Source
AI summary
Methods for improved parallel motion estimation are provided that decouple the merging candidate list derivation and motion estimation for merge mode and skip mode and the advanced motion vector predictor (AMVP) candidate list construction from regular motion estimation to increase the coding quality in parallel motion estimation while meeting throughput requirements. This decoupling may be accomplished by modifying the availability rules for spatial motion data (SMD) positions for construction of the candidate lists. As part of the decoupling, largest coding units (LCUs) of a picture may be divided into non-overlapping parallel motion estimation regions (PMER) of equal size. Within a PMER, motion estimation for merge mode, skip mode, and normal inter-prediction mode may be performed in parallel for all the prediction units (PUs) in the PMER.