Convex-Polyhedral Voxel Grid for High-Resolution Image Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image reconstruction techniques using 3D point clouds and voxel representations lack the necessary resolution for high-fidelity image generation, fail to capture fine details, and struggle with real-time scalability, eye-dependent lighting, and thin objects in dynamic environments.
Innovation Solution
A method involving capturing visible-light images from multiple viewpoints, dividing the 3D space into convex-polyhedral regions, and creating a 3D data structure to store image portions within these regions, enabling high-resolution image reconstruction with accurate representation of fine details and dynamic environments.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 3D point clouds and voxel representations are used for image reconstruction, then the data structures can be generated and utilized, but the resolution is insufficient for high-fidelity image generation and fine details cannot be captured
Solution Approach 1:
The patent divides the 3D space into multiple convex-polyhedral regions (voxels) and further segments each voxel into multiple sub-voxels. This hierarchical segmentation allows the system to capture fine details at higher resolutions while maintaining manageable data structure complexity through localized processing and storage of image portions in each sub-voxel.
2Adaptability or versatility
If 3D meshing is used for environment representation, then the data structure is suitable for static environments, but it cannot capture high detail geometry and is not suitable for dynamic environments
Solution Approach 1:
The patent employs a dynamic voxel-based data structure that can be updated and modified to represent changing environments. Unlike static 3D meshing, the voxel grid can accommodate dynamic objects and environmental changes by adding, removing, or modifying voxels as needed, enabling real-time adaptation to dynamic scenes while maintaining geometric detail through sub-voxel segmentation.
3Productivity
If existing data structures are used for image reconstruction, then the processing can be performed, but the quality is poor and real-time scaling to high quality is not achievable on existing hardware
Solution Approach 1:
By segmenting the 3D space into voxels and further into sub-voxels, the patent enables parallel processing of image portions across multiple hardware units. Each sub-voxel can be processed independently, allowing real-time scaling to high qualities by distributing the computational load across existing hardware resources while maintaining high resolution through the fine-grained segmentation structure.
4Reliability
If conventional image reconstruction techniques are used, then the basic reconstruction can be performed, but eye-dependent lighting such as specular highlights or reflections cannot be correctly captured and reproduced
Solution Approach 1:
The patent stores multiple image portions captured from different viewpoints within each sub-voxel, enabling local quality enhancement for specific visual effects. When reconstructing images, the system can selectively retrieve and composite image portions that capture eye-dependent lighting effects such as specular highlights and reflections, achieving accurate lighting reproduction by leveraging locally stored multi-view information without requiring complex global reconstruction techniques.
Data Source
AI summary
A computer-implemented method including: capturing visible-light images via visible-light camera(s) from view points in real-world environment, wherein 3D positions of view points are represented in coordinate system; dividing 3D space of real-world environment into 3D grid of convex-polyhedral regions; creating 3D data structure including nodes representing convex-polyhedral regions of 3D space; determining 3D positions of pixels of visible-light images based on 3D positions of view points; dividing each visible-light image into portions, wherein 3D positions of pixels of given portion of said visible-light image fall inside corresponding convex-polyhedral region; and storing, in each node, portions of visible-light images whose pixels' 3D positions fall inside corresponding convex-polyhedral region, wherein each portion of visible-light image is stored in corresponding node.