Affine Motion Prediction Using Non-Adjacent Candidates in Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding systems face challenges in reducing complexity and storage requirements for affine motion compensation, particularly with non-adjacent candidates, which require significant computational resources and memory for motion vector storage.
Innovation Solution
The proposed method generates affine merge and AMVP lists using non-adjacent affine candidates by deriving Control-Point Motion Vectors (CPMVs) from neighboring blocks outside the available region, replacing them with derived CPMVs, and using these to encode or decode video blocks, thereby reducing storage and computational demands.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If non-adjacent affine candidates are used for motion estimation, then motion compensation accuracy is improved, but storage requirements and computational complexity increase significantly
Solution Approach 1:
The patent segments the motion estimation process by dividing the search space into non-adjacent candidate blocks rather than using all neighboring blocks. This selective segmentation reduces the number of candidates while maintaining accuracy in regions with smooth motion characteristics.
Solution Approach 2:
The patent applies different motion estimation strategies to different regions based on their characteristics. Non-adjacent candidates are used in regions where motion is smooth and predictable, while traditional methods are applied where motion is complex, optimizing the balance between accuracy and complexity locally.
2Measurement precision
If non-adjacent affine candidates are used for motion estimation, then motion compensation accuracy is improved, but storage requirements increase
Solution Approach 1:
The patent extracts and stores only the essential motion information from non-adjacent candidate blocks rather than storing complete block data. This selective extraction reduces storage requirements while preserving the necessary motion compensation accuracy.
Solution Approach 2:
The patent creates a universal motion candidate list that serves multiple functions: it provides motion information for current blocks, enables prediction for future blocks, and reduces storage by reusing candidate data across multiple operations.
3Productivity
If non-adjacent affine candidates are used for motion estimation, then coding efficiency is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary motion estimation using non-adjacent candidates to establish initial motion vectors before performing more precise motion compensation. This preliminary action reduces the computational burden of subsequent processing steps.
Solution Approach 2:
The patent applies partial motion estimation to regions where full analysis is not necessary, using non-adjacent candidates only where needed to achieve sufficient accuracy without the full processing cost of traditional methods.
Data Source
AI summary
Methods and apparatus for video coding using non-adjacent affine candidates are provided. According to this method, one or more neighboring MVs (motion vectors) are determined from one or more non-adjacent affine-coded neighbors of the current block. CPMVs (Control-Point Motion Vectors) are determined based on said one or more neighboring MVs, wherein if a target neighboring block associated with one target neighboring MV (Motion Vector) is outside an available region, a derived CPMV) is generated to replace the target neighboring MV. An affine merge list or an affine AMVP (Advanced Motion Vector Prediction) list having said one or more neighboring MVs as one non-adjacent affine candidate is generated, wherein said one non-adjacent affine candidate generates a non-adjacent affine predictor using motion information according to the CPMVs. The current block is encoded or decoded using a motion candidate selected from the affine merge list or the affine AMVP list.