AR Object Recognition via Edge-Based Digital Model Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing augmented reality technologies face challenges in accurately and efficiently learning feature information of real objects, especially in dynamic environments, requiring high-level vision technology and repeated manufacturing of real models for new or partially changed objects.
Innovation Solution
A method involving edge detection from a digital model of a target object, classification of edges based on characteristics, and setting sample points, which generates object recognition library data for recognizing real objects. This method is performed using a computer-aided design program and enables fast detection and accurate tracking of a real object's pose.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If markerless tracking technology is used to recognize objects in augmented reality, then the application field is widened and markers are not required, but the accuracy of matching extraction deteriorates when environmental information changes
Solution Approach 1:
The patent segments the object recognition process into two parts: using markerless tracking for initial object location and acquisition, then applying template matching with pre-acquired image data for precise matching and extraction. This segmentation allows the system to benefit from both the versatility of markerless tracking and the accuracy of template-based methods.
Solution Approach 2:
The patent performs preliminary action by acquiring and storing template image data of the target object before the actual augmented reality application. This pre-acquired data serves as a reference for accurate matching during runtime, enabling the system to maintain high accuracy without requiring markers or complex environmental adaptations.
2Measurement precision
If deep learning methods are used to improve matching accuracy, then recognition accuracy is improved, but considerable effort and time are required to extract feature information
Solution Approach 1:
The patent performs feature extraction and template creation as a preliminary action during an offline setup phase. By pre-processing and storing the template data before runtime, the system avoids time-consuming feature extraction during actual augmented reality application, thus achieving both high accuracy and real-time performance.
Solution Approach 2:
The patent creates a copy of the target object's image data as a template, which can be repeatedly used for matching without requiring re-extraction of feature information. This copying approach eliminates the need for repeated deep learning feature extraction, significantly reducing computation time while maintaining recognition accuracy.
3Measurement precision
If traditional augmented reality methods using database comparison or marker tracking are used, then initial pose can be defined, but a lot of time and effort are required and markers are required which limits commercialization
Solution Approach 1:
The patent enables the system to automatically acquire and define the initial pose of the target object through markerless tracking, eliminating the need for manual pose definition by users. The system serves itself by automatically locating and tracking the object in the environment, thereby reducing both time and effort requirements while removing the dependency on markers.
Solution Approach 2:
The patent replaces the mechanical marker-based tracking system with a vision-based markerless tracking approach. This substitution eliminates the need for physical markers while maintaining the ability to define and track object pose, thereby reducing setup effort and enabling broader commercialization applications.
4Measurement precision
If repeated manufacturing of real models is performed to learn new or partially changed objects, then learning accuracy is improved, but manufacturing cost and time increase
Solution Approach 1:
The patent uses digital copying of object images through photography and image processing instead of physical reproduction through manufacturing. By capturing image data of the target object and creating digital templates, the system achieves accurate object learning without the need to manufacture physical models, thereby eliminating associated costs and time while maintaining high learning accuracy.
Solution Approach 2:
The patent replaces the mechanical manufacturing process with a digital imaging and processing approach. Instead of physically manufacturing reference models for learning, the system uses cameras to capture and process images, substituting mechanical production with optical and computational methods that are faster and more cost-effective.
Data Source
AI summary
A method of learning a target object by detecting an edge from a digital model of the target object and setting a sample point according to one embodiment of the present disclosure, which is performed by a computer-aided design program of an authoring computing device, includes: displaying a digital model of a target object that is a target of image recognition; detecting edges on the digital model of the target object; classifying the detected edges according to a plurality of characteristics; obtaining sample point information on the detected edges; and generating object recognition library data for recognizing a real object implementing the digital model of the target object based on the detected edges, characteristic information of the detected edges, and the sample point information.