Fast Cross Motion Estimation for Video Encoding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional motion estimation methods in video encoding, such as three-step search and diamond search, are inefficient and prone to trapping in local minimal values, leading to suboptimal performance in semiconductor devices.
Innovation Solution
A fast cross-motion estimation method that determines early termination based on a threshold value related to the quantized step length, uses previous frame motion vectors for prediction, and employs a 3×3 cross searching pattern to refine motion vector estimation, significantly reducing calculation and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional three-step search method is used, then motion estimation can be performed with a fixed searching pattern, but the search is trapped in local minimal values and becomes ineffective for small motion vectors
Solution Approach 1:
The patent applies dynamics by making the search pattern adaptive rather than fixed. The search pattern dynamically adjusts its shape and size based on the motion vector magnitude and search progress. For small motion vectors, the pattern becomes more focused to avoid local minima, while for large motion vectors, it expands to cover broader search areas. This dynamic adaptation resolves the contradiction between ease of operation and measurement precision.
Solution Approach 2:
The patent changes key parameters of the search pattern including its shape, size, and step length based on motion vector characteristics. The search pattern transforms from fixed geometric shapes to variable patterns that adapt to local image content and motion characteristics. This parameter changes approach allows the system to maintain operational simplicity while achieving high precision motion estimation.
2Reliability
If diamond search method uses large search pattern, then search scope is large and prevents trapping in local minimal, but search step is long and reduces search efficiency
Solution Approach 1:
The patent segments the motion estimation process into multiple phases with different search patterns. Instead of using a single large search pattern throughout, the method divides the search into coarse search phase (using larger patterns) and fine search phase (using smaller patterns). This segmentation allows the system to benefit from both large search scope and fast convergence, resolving the contradiction between reliability and productivity.
Solution Approach 2:
The search pattern dynamically adjusts its size and shape during the estimation process. The pattern starts larger to cover broad search areas and prevent local minima trapping, then progressively shrinks and refines as the search converges toward the optimal match point. This dynamic behavior maintains both reliability and productivity throughout the search process.
3Measurement precision
If diamond search method uses small search pattern, then precise location of optimal matching block is achieved, but search is trapped in local minimal
Solution Approach 1:
The patent applies preliminary action by performing a coarse search phase before the fine search phase. The preliminary coarse search uses larger search patterns to identify the general region of the optimal match point, ensuring the search does not start from a poor initial position that would lead to local minima. This preliminary action provides a reliable starting point for the subsequent precise search.
Solution Approach 2:
The search pattern dynamically transitions from larger to smaller sizes as the search progresses. Early in the search process, larger patterns provide robustness and prevent local minima trapping. As the search converges, the pattern shrinks to achieve precise localization. This dynamic adaptation resolves the contradiction between precision and reliability.
4Ease of manufacture
If conventional motion estimation methods are used, then encoding can be performed, but calculation complexity is high and encoder performance is limited
Solution Approach 1:
The patent applies partial action by performing motion estimation only where necessary and using simplified search patterns for regions with small motion vectors. Instead of applying complex search algorithms uniformly across all blocks, the method selectively applies different levels of search complexity based on local image characteristics and motion magnitude, reducing overall calculation complexity while maintaining encoding capability.
Solution Approach 2:
The patent changes search parameters including pattern size, step length, and search range based on motion vector magnitude and image content characteristics. This adaptive parameter adjustment reduces the number of search points and calculations required for blocks with small motions, while maintaining thorough search for blocks with large motions. This resolves the contradiction between ease of manufacture and productivity.
Data Source
AI summary
This disclosure is related to a fast motion estimation method for implementing in a video encoder. A particular method includes determining early termination based on the value of a sum of absolute difference (SAD0). The method also includes performing motion prediction using a motion vector from a previous frame as a prediction point. The method further includes using the prediction point as the center to obtain the integral motion vector by conducting a 3×3 cross searching until a minimum sum of absolute difference (minSAD) is located at the center of the cross and predicting a location of a half-pixel point based upon a point corresponding to the integral motion vector, values of the minimum sum of absolute difference of other four cross points.