AR Object Positioning via Virtual Frustum Point Cloud Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing augmented reality systems face limitations in accurately estimating the position and size of real-world objects due to the need for prior image/object scans and sensitivity to lighting conditions, which restricts their ability to recognize objects without pre-scanned data and limits processing power and communication speeds of mobile devices.
Innovation Solution
A method and system that utilize a camera and computer vision model to capture and process images, generate a bounding shape, construct a virtual frustum, identify points within the frustum, and calculate a representative distance to determine the position and size of real-world objects, improving accuracy and reducing reliance on pre-scanned data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image recognition systems use pre-scanned images for comparison, then recognition accuracy is improved, but the system cannot recognize objects without prior scans and requires similar lighting conditions
Solution Approach 1:
The patent introduces an intermediary geometric model (virtual frustum) that bridges the gap between 2D image data and 3D object properties. Instead of directly comparing images, the system uses the frustum as a mediator to project point cloud data into a standardized geometric representation that can be compared across different viewing conditions and lighting environments.
Solution Approach 2:
The patent transitions from 2D image space to 3D geometric space by constructing virtual frustums. This dimensionality change allows the system to represent objects in a coordinate system that is independent of lighting conditions and viewing angles, enabling recognition of objects without pre-scanned images while maintaining measurement precision.
2Adaptability or versatility
If mobile devices process complex AR computations, then AR functionality is improved, but processing power and communication speeds are limited
Solution Approach 1:
The patent extracts only the essential geometric features needed for AR functionality by constructing simplified virtual frustums from point cloud data. This extraction approach removes unnecessary computational complexity while retaining the core functionality needed for object positioning and measurement, making AR computations feasible on mobile devices with limited processing power.
Solution Approach 2:
The patent segments the complex task of object recognition and measurement into distinct steps: capturing point cloud data, constructing virtual frustums, and performing geometric comparisons. This segmentation allows each step to be optimized independently and reduces the overall computational burden on mobile devices.
3Adaptability or versatility
If AR systems overlay virtual elements on real-world scenes, then seamless mixing of virtual and real-world elements is improved, but accurate position and size estimation of real-world objects becomes challenging
Solution Approach 1:
The patent changes the parameters used for object representation from pixel-based 2D image coordinates to 3D geometric parameters defined by virtual frustums. This parameter transformation enables accurate position and size estimation by using geometric relationships that are invariant to lighting conditions and camera parameters, thereby improving measurement precision while maintaining seamless AR integration.
Data Source
AI summary
The disclosure relates to systems, methods and processor-readable storage mediums, having processor-executable instructions stored thereon, for determining the position of a real-world object in an augmented reality application running on a computing device having a camera. The method includes capturing image information using the camera. Detection of real-world objects is then performed using a computer vision model, and filtering of a received point cloud is performed by way of a virtual frustum constructed using the captured image information. Finally, the position of the object is determined using the filtered points.