Image Motion Vector Prediction Through Static-Dynamic Mesh Separation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current frame prediction technologies require high computing power and power consumption due to the need for pixel-by-pixel color or brightness matching to determine object positions in different frame images, which complicates the prediction of future frames.
Innovation Solution
An image processing method that separates the calculation of motion vectors for static and dynamic meshes, using intermediate rendering variables such as MVP matrices and depth data, and coordinate data to predict future frames efficiently, reducing computational overheads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If pixel-by-pixel color or brightness matching is used to determine object positions in different frame images, then prediction accuracy is improved, but computing power requirements and power consumption increase significantly
Solution Approach 1:
The patent segments the image processing task by separating static meshes from dynamic meshes. Static meshes are processed using MVP matrices and depth data from intermediate rendering variables, while dynamic meshes use coordinate data. This segmentation allows each type to be processed with appropriate methods, reducing overall computational complexity while maintaining accuracy.
Solution Approach 2:
The patent extracts the motion vector calculation for static meshes from the pixel-by-pixel matching process. By using intermediate rendering variables (MVP matrices and depth data) that are already available during rendering, the patent extracts and utilizes this information to determine static mesh motion vectors without requiring additional pixel-level analysis.
2Measurement precision
If pixel-by-pixel color or brightness matching is performed for each pixel point, then object position determination accuracy is improved, but power consumption increases
Solution Approach 1:
The patent performs preliminary action by capturing and storing intermediate rendering variables (MVP matrices and depth data) during the rendering process itself. These variables are prepared in advance and used later for motion vector calculation, eliminating the need for energy-intensive pixel-by-pixel matching during the prediction phase.
Solution Approach 2:
The patent introduces intermediate rendering variables as mediators between the rendering process and the motion vector calculation. These variables (MVP matrices and depth data) serve as efficient intermediaries that convey necessary spatial information without requiring direct pixel-level analysis, thus reducing power consumption.
3Power
If separate calculation of static mesh and dynamic mesh motion vectors is implemented, then computing power overheads are reduced, but implementation complexity increases
Solution Approach 1:
The patent achieves multi-functionality by utilizing the existing rendering pipeline to serve dual purposes: both rendering the image and capturing intermediate rendering variables for later motion vector calculation. The same rendering process that creates the visual output also prepares the data needed for efficient static mesh motion analysis, reducing overall system complexity.
Data Source
AI summary
Embodiments of this application relate to the field of image processing, and disclose an image processing method and an electronic device, which can accurately and quickly calculate motion vectors of a static mesh and a dynamic mesh separately, thereby reducing computing power overheads for vector calculation, and improving prediction efficiency. A specific solution includes: determining a position of a static mesh of a next frame image based on intermediate rendering variables of at least two completed frame images, where each intermediate rendering variable includes an MVP matrix and depth data of a corresponding frame image; and determining, based on coordinate data of a first model in the at least two completed frame images, a position of the first model in the next frame image. A mesh corresponding to the first model is a dynamic mesh in the image.