Multi-layer Reprojection for AR Registration Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing reprojection techniques for virtual content in augmented reality (AR) applications fail to effectively address registration errors caused by changes in user perspective over time, especially when virtual objects are registered to different points in the real world at varying depths.
Innovation Solution
The approach involves identifying virtual objects in a single layer eye buffer and associating them with multiple layers based on different depth values. A reprojection engine performs per-layer reprojection on these virtual objects, rather than single-layer reprojection, to minimize registration errors. This is achieved by defining bounding boxes around the virtual objects and applying metadata indicative of edge information and reprojection instructions for each bounding box.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a single homography is applied to warp all virtual objects at varying depths, then the processing complexity is reduced, but registration errors increase for virtual objects at different depths
Solution Approach 1:
The patent segments virtual objects into multiple depth layers based on their registered depth values. Each layer is then processed with its own homography transformation, allowing different warp factors for objects at different depths. This segmentation resolves the contradiction by maintaining registration accuracy for multi-depth objects while managing processing complexity through hierarchical organization.
Solution Approach 2:
The patent applies local quality by assigning different homography transformations to different depth layers. Instead of a uniform transformation for all objects, each layer receives customized warp parameters based on its depth characteristics. This allows optimal registration for each depth level while organizing the complexity through structured layer-based processing.
2Measurement precision
If per-layer reprojection is performed for virtual objects at different depths, then registration accuracy is improved, but processing time increases
Solution Approach 1:
By segmenting objects into depth layers, the system can process each layer independently and efficiently. The segmentation enables parallel processing of different depth groups, reducing overall processing time while maintaining high registration accuracy for each layer through customized homography transformations.
Solution Approach 2:
The patent performs preliminary actions by pre-calculating and storing depth information and homography parameters for different layers before the actual reprojection. This preparation allows the reprojection process to proceed more quickly using pre-computed data, reducing processing time while maintaining accuracy.
3Reliability
If virtual content is reprojected to correct perspective changes, then registration errors are reduced, but computational resources are consumed
Solution Approach 1:
The patent segments the reprojection process into manageable depth layers, each processed with appropriate computational resources. This segmentation allows the system to apply reprojection only where needed and at appropriate complexity levels, maintaining registration stability while optimizing computational resource consumption.
Solution Approach 2:
The patent changes parameters such as homography matrix complexity and warp factor based on depth layer characteristics. Closer objects receive more aggressive warping with higher computational cost, while distant objects use simpler transformations. This parameter adaptation maintains registration stability while reducing overall computational resource consumption.
Data Source
AI summary
This disclosure provides systems, devices, apparatus and methods, including computer programs encoded on storage media, for multi-layer reprojection techniques for augmented reality. A display processor may obtain a layer of graphics data including a plurality of virtual objects. Each the plurality of virtual objects may be associated with at least one bounding box of a plurality of bounding boxes. The display processor may further obtain metadata indicative of at least one edge of the at least one bounding box of the plurality of bounding boxes, and metadata corresponding to reprojection instructions associated with each of the plurality of bounding boxes. The display processor may reproject the plurality of virtual objects based on the metadata indicative of the at least one edge of the at least one bounding box and the metadata corresponding to the reprojection instructions.