Motion Vector Refinement Using Integer Rounding

Solution Overview

Problem

Motion vector refinement in video encoding and decoding requires increased on-chip memory or external memory access, leading to performance slowdowns, especially when initial motion vectors point to fractional positions, necessitating interpolation of multiple samples which can be costly in terms of memory access.

Innovation Solution

Rounding initial motion vector coordinates to integer positions reduces the complexity of interpolation and memory access by limiting the search space to integer sample positions, thereby reducing the number of fractional positions and samples needed for interpolation.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If motion vector refinement is performed with fractional sample positions, then prediction accuracy is improved, but memory access requirements and computational complexity increase

Engineering Contradiction:

Improveprediction accuracyVSAvoidmemory access requirements

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the motion vector refinement process into two distinct stages: first performing refinement on integer sample positions, then optionally performing a second refinement pass on fractional sample positions. This segmentation allows the system to balance between prediction accuracy and memory access requirements by enabling selective application of computationally intensive fractional refinement only where needed.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by performing motion vector refinement only on a subset of motion vectors rather than all motion vectors in the picture. The system identifies candidate motion vectors that benefit most from refinement and applies the computationally intensive fractional sample position refinement selectively to these candidates, rather than uniformly to all motion vectors.

Inventive Principle:

Principle #16Partial or excessive action

2Measurement precision

If motion vector refinement is performed with fractional sample positions, then prediction accuracy is improved, but processing speed decreases

Engineering Contradiction:

Improveprediction accuracyVSAvoidprocessing speed

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The patent segments the motion vector refinement process into two distinct stages: first performing refinement on integer sample positions, then optionally performing a second refinement pass on fractional sample positions. This segmentation allows the system to balance between prediction accuracy and memory access requirements by enabling selective application of computationally intensive fractional refinement only where needed.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by performing motion vector refinement only on a subset of motion vectors rather than all motion vectors in the picture. The system identifies candidate motion vectors that benefit most from refinement and applies the computationally intensive fractional sample position refinement selectively to these candidates, rather than uniformly to all motion vectors.

Inventive Principle:

Principle #16Partial or excessive action

3Measurement precision

If search space includes fractional positions, then motion vector precision is improved, but on-chip memory size requirements increase

Engineering Contradiction:

Improvemotion vector precisionVSAvoidon-chip memory size

Core Design Contradiction:

Measurement precisionVSVolume of stationary object

Solution Approach 1:

The patent segments the motion vector refinement process into two distinct stages: first performing refinement on integer sample positions, then optionally performing a second refinement pass on fractional sample positions. This segmentation allows the system to balance between prediction accuracy and memory access requirements by enabling selective application of computationally intensive fractional refinement only where needed.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by performing motion vector refinement only on a subset of motion vectors rather than all motion vectors in the picture. The system identifies candidate motion vectors that benefit most from refinement and applies the computationally intensive fractional sample position refinement selectively to these candidates, rather than uniformly to all motion vectors.

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS11159820B2Motion vector refinement of a motion vector pointing to a fractional sample position

Publication Date: 2021.10.26 HUAWEI TECH CO LTD

AI summary

A motion vector determination employs template matching. At first, an initial motion vector is obtained. If the initial motion vector points to a position that is fractional, and thus, requires interpolation from the integer sample positions, the initial motion vector is rounded to a closest integer sample position. The rounded position is then used to define the search space for the template matching to define a refinement of the initial motion vector.