Affine Motion Vector Prediction for Video Coding Efficiency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video coding systems face challenges in efficiently handling complex motion within video sequences, particularly in affine motion models, which are not adequately addressed by current motion vector prediction techniques.
Innovation Solution
The implementation of affine motion vector prediction (MVP) techniques, including the generation of affine MVP candidate lists, which derive motion vectors using inherited and constructed candidates from neighboring blocks coded with affine motion models, to enhance coding performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional motion vector prediction techniques are used, then coding simplicity is maintained, but coding efficiency deteriorates when handling complex motion in video sequences
Solution Approach 1:
The patent segments the motion model into two parts: a simple translational component (traditional MV) and an affine component (handling complex motion). This segmentation allows the system to maintain basic coding efficiency while adding affine parameters only where complex motion is detected, thus improving overall coding efficiency without uniformly increasing complexity across all blocks.
Solution Approach 2:
The patent dynamically selects between traditional motion vector prediction and affine motion vector prediction based on the complexity of motion in each block. The affine mode is activated when complex motion is detected, allowing the system to adapt its complexity level to the actual motion characteristics, thereby improving coding efficiency for complex motion regions while maintaining simplicity for regular motion regions.
2Reliability
If affine motion models are implemented, then handling of complex motion is improved, but computational complexity increases
Solution Approach 1:
The patent applies affine motion models only to specific blocks where complex motion is detected, rather than uniformly across the entire video sequence. This local application of the more complex affine model improves motion representation accuracy where needed while avoiding unnecessary computational complexity in regions where traditional motion models suffice.
Solution Approach 2:
The patent changes the number of motion parameters from the traditional two-component MV to a four-component affine MV plus two MVDs when complex motion is detected. This parameter change improves motion representation accuracy for complex motion patterns while being selectively applied only when necessary, thus managing computational complexity.
3Measurement precision
If more motion vector predictors are generated, then prediction accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary detection of complex motion characteristics before generating motion vector predictors. By identifying blocks with complex motion in advance, the system can selectively generate the more computationally intensive affine MVP candidates only for those blocks, thereby improving prediction accuracy where needed while minimizing the overall processing time increase.
Solution Approach 2:
The patent generates affine MVP candidates partially - only for blocks where complex motion is detected - rather than for all blocks. This partial application of the more thorough predictor generation process improves motion vector prediction accuracy for complex motion regions while avoiding the full processing time cost of applying it universally.
Data Source
Figure 1A~1B
Figure 2~3
Figure 4A~4B
AI summary
Methods and apparatus of Inter prediction for video coding including affine Inter mode are disclosed. In one method, an affine MVP candidate list is generated, wherein the affine MVP candidate list comprises at least one inherited affine MVP derived from a neighbouring block set. Prediction differences of a current controlpoint MV set associated with the affine motion model are encoded using one predictor selected from the affine MVP candidate list at the video encoder side or the prediction differences of the current controlpoint MV set associated with the affine motion model are decoded at the video decoder side using one predictor selected from the affine MVP candidate list. In another method, the inherited affine MVP is derived by considering whether at least one of reference picture lists of said at least one neighbouring block includes one reference picture being the same as a current reference picture of the current block.