Image Depth Map Generation via Global Motion Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional 2D-to-3D conversion techniques, such as computer vision-based methods and global depth models, are not practical for real-time operation and are inaccurate in representing local depth discontinuities, especially under poor lighting conditions and with global motion changes.
Innovation Solution
The system generates image depth maps using a full global depth map adapted to image global motion and localized depth analysis, utilizing relative relationships of pixel attributes across depth discontinuities, which includes tracking global motion and updating depth maps based on saliency area detection to create a more accurate representation of local depth.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If computer vision based technologies are used for 2D-to-3D conversion, then depth information can be extracted, but significant computing resources are required and real-time operation is not achieved
Solution Approach 1:
The patent segments the depth map generation process into global depth model generation and local depth refinement stages. The global depth model provides a coarse depth estimation that can be computed quickly, while local refinement techniques are applied only to specific regions requiring higher precision. This segmentation allows the system to achieve real-time performance while maintaining accurate depth extraction.
Solution Approach 2:
The patent applies local quality by using localized depth analysis techniques that focus computational resources on regions with depth discontinuities or significant features, rather than processing the entire image uniformly. This allows real-time operation by concentrating computing power where it is most needed while using simpler models for other regions.
2Productivity
If global depth models are used for 2D-to-3D conversion, then processing speed can be improved, but local depth discontinuities are not accurately represented
Solution Approach 1:
The patent divides the depth map into global and local components. The global depth model is generated using efficient algorithms for fast processing, while local depth refinement is applied to specific regions to capture depth discontinuities. This segmentation enables both fast processing and accurate local depth representation.
Solution Approach 2:
The patent merges the global depth model with local depth refinement results to produce the final depth map. The global model provides the overall depth structure and processing speed, while local refinement adds accuracy for depth discontinuities. This combination achieves both fast processing and precise local depth representation.
3Adaptability or versatility
If centralized depth models are used, then a generalized depth representation can be provided, but the model cannot dynamically represent scene changes and motion
Solution Approach 1:
The patent implements dynamics by making the depth model adaptive to scene changes and motion. The global depth model is updated based on detected scene changes, and local depth refinement is applied dynamically to regions with motion or depth discontinuities. This allows the model to represent changing scenes while maintaining consistency through structured update mechanisms.
Solution Approach 2:
The patent uses feedback mechanisms where the depth model is continuously refined based on scene analysis results. Motion detection and scene change detection provide feedback that triggers updates to the global depth model and activates local refinement in relevant regions, enabling dynamic adaptation while maintaining model stability through controlled update processes.
Data Source
AI summary
Systems and methods which provide generation of image depth maps which more accurately represent the local depth discontinuity within images through use of image global depth maps adapted based upon image global motion and/or localized depth analysis utilizing relative relationships of attributes across depth discontinuities in the image are disclosed. Embodiments utilize a full global depth map which is larger than or equal to the image being converted in order to accommodate image global motion, in generating an image global depth map. In operation according to embodiments, an image global depth map is identified within the full global depth map, such as based upon global motion within the image. Localized depth analysis, using pixel attribute relative relationships, is applied with respect to the image global depth map according to embodiments to generate an image depth map which more accurately reflects the local depth discontinuities within the image.