Motion Vector Prediction for Multi-View Video Coding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing motion vector prediction techniques for multi-view video coding face inefficiencies, particularly when handling motion vectors that point to frames with the same picture order count but in different views, leading to suboptimal compression performance and issues with scaling factors becoming zero or indefinite.
Innovation Solution
The method categorizes motion vectors into different types and restricts candidate predictor selection based on these types, determining the motion vector type based on the current and reference camera views and their respective points in time, and uses a scaling factor formula that avoids zero or indefinite values by setting it to a fixed value in special cases.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If motion vector prediction is performed without categorizing motion vector types in multi-view video coding, then all candidate predictors can be considered, but the complexity increases and coding efficiency deteriorates due to suboptimal predictor selection
Solution Approach 1:
The patent segments motion vectors into different types (temporal, inter-view, and combined motion vectors) based on their reference frame relationships. This segmentation allows the encoder to select appropriate candidate predictors for each motion vector type, improving coding efficiency by avoiding suboptimal predictor selection while managing complexity through structured categorization.
2Reliability
If scaling factor is calculated using the formula with picture order count differences for motion vectors pointing to frames with the same POC in different views, then temporal scaling is applied, but the scaling factor becomes zero or indefinite
Solution Approach 1:
The patent applies preliminary anti-action by detecting the special case where reference frames have the same picture order count before calculating the scaling factor. When this condition is detected, the patent preemptively sets the scaling factor to a fixed value (typically 1) instead of applying the standard formula, thereby preventing the zero or indefinite scaling factor problem before it occurs.
3Manufacturing precision
If motion vectors from all views and time points are considered as candidate predictors, then more prediction options are available, but the bitrate consumption increases due to more index signaling
Solution Approach 1:
The patent segments the candidate predictor pool into motion vector type-specific candidates. For each motion vector type (temporal, inter-view, combined), only relevant candidate predictors are considered. This segmentation maintains prediction accuracy by selecting appropriate candidates while reducing bitrate consumption by limiting the number of index signals that need to be transmitted.
Solution Approach 2:
The patent applies local quality by tailoring the candidate predictor selection to the specific characteristics of each motion vector type. Different motion vector types have different local prediction qualities from different candidate pools, and the patent optimizes predictor selection for each type locally rather than using a uniform approach, improving overall prediction accuracy while managing bitrate.
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
The embodiments relates to motion vector prediction and decoding for multi-view video content from multiple camera views (10, 20). At least one candidate motion vector predictor (44, 54) is provided for a current motion vector (34) to be encoded. A motion vector predictor (44) is then determined from the at least one candidate motion vector predictor (44, 54). Correspondingly, during decoding a motion vector (34) is determined for a current pixel block (30) based on a motion vector predictor (44) determined for the current pixel block (30).