Motion Vector Refinement With a Limited Memory Access Window
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current motion vector refinement methods require increased on-chip memory size or external memory access, which is undesirable due to processing slowdowns.
Innovation Solution
Limit the number of samples accessible for motion vector refinement by defining a window around the initial motion vector, allowing only integer samples within this window for interpolation filtering, and iteratively extending the search space based on best matching positions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If motion vector refinement is performed using template matching with interpolation filtering, then motion vector accuracy is improved, but on-chip memory size requirements increase
Solution Approach 1:
The patent segments the reference picture into multiple blocks and processes template matching in a limited window around the initial motion vector position rather than accessing the entire reference picture. This segmentation reduces the memory footprint required for motion vector refinement while maintaining accuracy in the critical search region.
Solution Approach 2:
The patent applies interpolation filtering and template matching only within a localized window around the initial motion vector position, rather than uniformly across the entire reference picture. This local processing approach concentrates computational resources where they are most needed, improving motion vector accuracy without proportionally increasing memory requirements.
2Measurement precision
If motion vector refinement uses a larger search space with fractional sample positions, then motion vector precision is improved, but external memory access increases causing processing slowdown
Solution Approach 1:
The patent divides the search space into a limited window around the initial motion vector position and processes only fractional sample positions within this window. This segmentation allows high-precision sub-pixel search without requiring access to the entire reference picture, thus avoiding external memory access bottlenecks.
Solution Approach 2:
The patent performs integer-sample template matching first to obtain an initial motion vector, then uses this initial vector to define a limited window for subsequent fractional-sample refinement. This preliminary action reduces the search space for the precision-enhancing fractional interpolation, thereby reducing memory access requirements and improving processing speed.
3Measurement precision
If interpolation filtering with predefined tap-size is applied to fractional sample positions, then sub-pixel accuracy is improved, but the number of accessible integer samples within the window is reduced
Solution Approach 1:
The patent adjusts the window size and position parameters based on the initial motion vector and the required tap-size for interpolation filtering. By dynamically parameterizing the window to match the specific refinement needs, the system ensures sufficient integer samples are available for accurate fractional position calculation without unnecessarily increasing memory access.
Data Source
AI summary
The present disclosure relates to motion vector refinement. As a first step, an initial motion vector and a template for the block are obtained. Then, the refinement of the initial motion vector is determined by template matching with said template in a search space. The search space is located on a position given by the initial motion vector and includes one or more fractional sample positions, wherein each of the fractional sample positions belonging to the search space is obtained by interpolation filtering with a filter of a predefined tap-size assessing integer samples only within a window, said window being formed by integer samples accessible for the template matching in said search space.