Augmented Reality Image Re-projection for Computational Load Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Augmented-reality devices face high computational expenses due to continuous rendering of augmentation imagery from three-dimensional models, which hampers their ability to provide efficient and performant mixed-reality experiences.
Innovation Solution
Implementing post-rendering re-projection techniques to transform two-dimensional augmentation images from a first virtual perspective into an updated image that approximates a second virtual perspective without additional rendering from the three-dimensional model, reducing the number of image rendering operations and computational load.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If continuous rendering from three-dimensional models is performed to provide augmented-reality experiences, then the quality and realism of augmentation imagery is improved, but computational expenses and processing time increase
Solution Approach 1:
The system pre-renders augmentation imagery from three-dimensional models at specific viewpoints before runtime. These pre-rendered images are stored and later transformed using re-projection techniques when the device views change, eliminating the need for continuous real-time rendering from 3D models and significantly reducing computational expenses.
Solution Approach 2:
Instead of continuously rendering from 3D models, the system creates copies of pre-rendered augmentation images and applies geometric transformations (re-projection) to these copies. This allows the system to generate updated augmentation imagery for different viewpoints by transforming existing 2D images rather than re-rendering from 3D models, thereby improving computational efficiency.
2Reliability
If continuous rendering operations are performed to update augmentation imagery, then the realism and immersion of mixed-reality experience is improved, but processing time and latency increase
Solution Approach 1:
Augmentation imagery is pre-rendered from three-dimensional models at designated viewpoints before the user actually views them. This preliminary rendering allows the system to prepare high-quality images in advance, reducing real-time processing requirements and latency when the user interacts with the augmented reality experience.
Solution Approach 2:
The system replaces the mechanical rendering process (continuous real-time 3D model rendering) with a computational transformation process (2D image re-projection). This substitution uses mathematical transformations to generate updated augmentation images from pre-rendered 2D images, significantly reducing processing time and latency while maintaining visual quality.
3Adaptability or versatility
If multiple rendering operations are performed for different virtual perspectives, then the adaptability of augmentation imagery to user movement is improved, but computational load and processing complexity increase
Solution Approach 1:
The system pre-renders augmentation imagery from multiple predetermined viewpoints and stores these images. When the user's perspective changes, the system selects the appropriate pre-rendered image and applies re-projection transformations rather than performing new rendering operations, reducing computational complexity while maintaining adaptability to user movement.
Solution Approach 2:
Instead of rendering new images for each user perspective change, the system creates transformed copies of existing pre-rendered 2D augmentation images. By applying geometric transformations to these copies, the system efficiently adapts augmentation imagery to different viewpoints without the high computational cost of repeated 3D model rendering.
Data Source
AI summary
A two-dimensional augmentation image is rendered from a three-dimensional model from a first virtual perspective. A transformation is applied to the augmentation image to yield an updated two-dimensional augmentation image that approximates a second virtual perspective of the three-dimensional model without additional rendering from the three-dimensional model.