AR Object Superimposition via Region Segmentation and Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current augmented reality technologies face challenges in efficiently superimposing virtual information onto real-world environments, requiring improved methods for identifying and processing object areas within video frames to accurately combine virtual and real-world elements.
Innovation Solution
A processing method utilizing a box selection model and classification model to determine object areas and categories within video frames, allowing for the precise superimposition of augmented reality scene information, such as 3D models and audio effects, by analyzing feature points and calculating similarity thresholds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If full-image analysis is performed to identify objects for AR superimposition, then measurement precision and reliability are improved, but processing time and productivity deteriorate
Solution Approach 1:
The patent divides the video frame processing into multiple stages: first identifying candidate regions using simplified criteria (motion detection, edge detection), then performing detailed object recognition only on these candidate regions. This segmentation approach maintains identification accuracy while reducing the overall processing load and time required for full-image analysis.
Solution Approach 2:
The patent applies partial action by performing complete object recognition only on identified candidate regions rather than the entire image. The candidate region identification uses lighter processing, while detailed analysis is concentrated on specific areas, achieving high precision where needed without the computational cost of full-image processing.
2Measurement precision
If complex object recognition algorithms are used to accurately identify target objects, then measurement precision is improved, but device complexity increases
Solution Approach 1:
The patent segments the recognition process into two distinct phases: candidate region detection using simple geometric and motion-based criteria, followed by category classification using machine learning models only on these candidates. This reduces overall algorithmic complexity while maintaining high classification accuracy through focused application of complex models.
Solution Approach 2:
Complex classification algorithms are applied partially only to candidate regions rather than the entire image. This selective application maintains high object classification accuracy while significantly reducing computational complexity and resource requirements compared to full-image classification.
3Reliability
If AR scene information is superimposed on every detected object, then completeness and reliability are improved, but processing time and system complexity increase
Solution Approach 1:
The patent applies different processing qualities to different regions: candidate region identification receives focused, detailed processing to ensure high reliability for potential AR targets, while other areas receive minimal or no processing. This local quality differentiation ensures AR scene completeness for relevant objects while reducing overall processing time.
Solution Approach 2:
The patent performs partial superimposition by applying AR scene information only to confidently identified target objects within candidate regions rather than attempting to process every possible object in the entire image. This maintains reliability for detected targets while significantly reducing processing time and system complexity.
Data Source
AI summary
A processing method that is performed by one or more processor is provided. The processing method includes determining a target video frame in a currently captured video; determining an object area in the target video frame based on a box selection model; determining a category of a target object in the object area based on a classification model used to classify an object in the object area; obtaining augmented reality scene information associated with the category of the target object; and performing augmented reality processing on the object area in the target video frame and the augmented reality scene information, to obtain the augmented reality scene.