Blurred 2.5D Representation Data for AR Object Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current object detection algorithms for augmented reality (AR) systems require time-consuming manual training processes and may not accurately reflect how objects appear to AR system sensors, leading to increased detection failures and computational power usage.
Innovation Solution
A method involving the generation of blurred 2.5D representation data from a 3D model, based on depth sensor characteristics, to create training data for object detection algorithms, which reduces the need for manual image capturing and improves detection accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual training methods are used where a trainer positions the object and captures numerous images from numerous different angles, then the training data can be obtained, but the training process becomes time-consuming and labor-intensive
Solution Approach 1:
The patent uses a 3D model (digital copy) of the object to generate training images instead of requiring physical objects and manual photographing. The 3D model can be virtually positioned and viewed from any angle, eliminating the need for physical repositioning and manual image capture while maintaining detection accuracy.
Solution Approach 2:
The system performs preliminary actions by pre-generating a comprehensive set of training images from the 3D model covering all possible angles and positions. This preliminary generation of training data eliminates the need for time-consuming manual capture during the actual training process.
2Ease of manufacture
If images of the object are used for training that do not accurately reflect how the object will appear to the AR system's sensors, then the training process is simpler, but detection failures increase and computational power is wasted
Solution Approach 1:
The patent applies sensor characteristic data (blurring parameters, noise characteristics, resolution limits) to transform the ideal 3D model into images that accurately reflect how the AR system's sensors will actually perceive the object. This parameter-based transformation ensures training images match real sensor input without manual adjustment.
Solution Approach 2:
The system introduces sensor characteristic data as an intermediary element between the 3D model and the training images. This intermediary transforms the perfect digital model into realistic sensor-like images, bridging the gap between idealized training data and actual sensor perception.
3Reliability
If comprehensive training data covering all angles and positions is obtained manually, then detection accuracy improves, but the complexity of the training process increases
Solution Approach 1:
The patent replaces the mechanical manual process of positioning objects and capturing images with an automated computational system. A processor automatically generates images from a 3D model by applying virtual camera positions and sensor characteristics, eliminating the need for physical manipulation and manual coordination.
Data Source
AI summary
A non-transitory computer readable medium embodies instructions that cause one or more processors to perform a method. The method includes: (A) receiving, in one or more memories, a 3D model corresponding to an object, and (B) setting a depth sensor characteristic data set for a depth sensor for use in detecting a pose of the object in a real scene. The method also includes (C) generating blurred 2.5D representation data of the 3D model for at least one view around the 3D model based on the 3D model and the depth sensor characteristic data set, to generate, on the basis of the 2.5D representation data, training data for training an object detection algorithm, and (D) storing the training data in one or more memories.