Parallel Rate-Constrained Motion Estimation Using Temporal Predictors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current video coding technologies, such as H.265/HEVC, face challenges in achieving high parallelization and rate-distortion (RD) performance due to high computational complexity, especially in rate-constrained motion estimation (RCME), which is exacerbated by the need for accurate motion vector predictors (MVPs) and high bandwidth usage.
Innovation Solution
The method involves performing RCME in two stages using multiple temporal motion vector predictors, determining Motion Vector Predictor Candidates (MVPCs), and selecting the optimal motion vector based on rate distortion optimization, suitable for GPU/CPU heterogeneous architectures, allowing for high parallelization and improved RD performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If prior art methods process RCME in parallel on several PUs by estimating MVPs from neighboring CTUs using spatial information, then parallelization degree is increased, but RD performance deteriorates due to inaccurate MVP prediction
Solution Approach 1:
The patent transitions from spatial information (neighboring CTUs in the same frame) to temporal information (collocated CTUs in reference frames) for MVP estimation. This dimensional shift allows parallel processing while maintaining accurate motion prediction, as temporal motion vectors from reference frames are already computed and can be reused without compromising RD performance.
Solution Approach 2:
The patent uses motion vectors from previously encoded reference frames (temporal MVPs) as predictors for the current frame. This preliminary action of reusing already-computed motion information from reference frames eliminates the need for complex spatial MVP estimation during parallel processing, thereby maintaining both high parallelization and accurate RD performance.
2Manufacturing precision
If a GPU is used as pre-processor to calculate SADs for the whole search region, then RD performance is improved, but bandwidth usage increases significantly reducing speedup
Solution Approach 1:
The patent extracts only the necessary motion vector information from reference frames (temporal MVPs) and uses it for parallel RCME processing on the GPU. Instead of transferring the entire search region data back to the CPU, only the essential motion vector candidates are utilized, significantly reducing bandwidth requirements while maintaining RD performance.
Solution Approach 2:
The patent copies motion vector information from reference frames to the current frame processing. By reusing the motion vectors already computed in reference frames as temporal predictors, the system avoids redundant calculations and excessive data transfer, achieving efficient parallel processing with reduced bandwidth consumption.
3Productivity
If HEVC uses hybrid architecture with numerous improvements in frame splitting, prediction modes, transformation, and filtering, then compression ratio is doubled, but computational complexity increases considerably
Solution Approach 1:
The patent segments the motion estimation process into independent parallel units by utilizing temporal MVPs from reference frames. Each PU can be processed independently using pre-computed temporal predictors, allowing the computational complexity to be distributed across multiple processing units simultaneously, thereby maintaining high compression ratios while reducing the burden on single processors.
Solution Approach 2:
The patent enables the encoding process to serve itself by reusing motion vectors from reference frames as predictors for the current frame. This self-service approach eliminates the need for complex inter-PU dependencies and iterative MVP estimation, significantly reducing computational complexity while preserving the high compression efficiency of HEVC's hybrid architecture.
Data Source
AI summary
Methods and systems for parallel rate-constrained motion estimation in a video encoder are provided. Embodiments of the present invention provide a solution for the problem of the reliance upon spatial dependencies when processing parallel RCME on a frame. In order to solve this problem, embodiments of the present invention determine a list of at least two Motion Vector Predictor Candidates (MVPC), determine, in parallel for at least two of said at least two MVPCs, corresponding Motion Vector Candidates (MVC) using Rate-Constrained Motion Estimation (RCME) and determine the optimal Motion Vector (MV) among the determined MVCs based on rate distortion optimization.