3D Model Deficiency Detection via Depth-Foreground Comparison
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating three-dimensional models from multi-view images often result in deficiencies due to blind spots or inaccuracies, leading to incomplete or inaccurate information, especially when multiple subjects are present or occlusion occurs, making it difficult to provide accurate free viewpoint images.
Innovation Solution
A method that analyzes the three-dimensional model by comparing depth and foreground images to identify deficiencies, compensates for these by generating new three-dimensional points, and outputs a composite image indicating the presence of deficiencies, ensuring accurate representation and notification to users.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If three-dimensional models are generated from multi-view images using conventional methods, then free viewpoint images can be generated, but deficiencies such as blind spots and inaccuracies occur leading to incomplete information
Solution Approach 1:
The patent implements a feedback mechanism by comparing depth images with foreground images to detect deficiencies in the three-dimensional model. The system identifies blind spots and inaccuracies where depth information is missing or incorrect, then uses this feedback to generate corrected three-dimensional models by filling in missing data from multiple viewpoint images, thereby improving reliability while maintaining free viewpoint generation capability
Solution Approach 2:
The patent performs preliminary analysis by generating depth images from the three-dimensional model and comparing them with foreground images extracted from multi-view images before finalizing the free viewpoint images. This preliminary detection of deficiencies allows the system to correct errors in advance, ensuring accurate representation while maintaining the versatility of free viewpoint generation
2Quantity of substance
If multiple subjects are present or occlusion occurs in the scene, then rich scene information is captured, but deficiencies in three-dimensional model appear making accurate representation difficult
Solution Approach 1:
The patent segments the scene into multiple subjects and regions by extracting foreground images from multi-view images. It then processes each segment separately to generate depth information and identify deficiencies in the three-dimensional model for each subject, allowing accurate representation even when multiple subjects are present or occlusion occurs, while maintaining comprehensive scene information
3Productivity
If conventional three-dimensional model generation is used without analysis, then processing is simpler and faster, but inaccurate information is provided due to blind spots and deficiencies
Solution Approach 1:
The patent performs preliminary generation of depth images and foreground images quickly from the three-dimensional model and multi-view images, then compares them to identify deficiencies. This preliminary action allows the system to maintain high processing speed while detecting and correcting information loss, ensuring both productivity and completeness of spatial information through efficient parallel processing
Data Source
AI summary
A method for analyzing a three-dimensional model of an object which includes: obtaining the three-dimensional model generated based on images of the object, the images being imaged by respective cameras from respective viewpoints, the three-dimensional model including three-dimensional points each of which indicating a position of the object; obtaining a camera parameter of one camera among the respective cameras; generating, based on the camera parameter and the three-dimensional model, a depth image indicating a distance between the one camera and the object; generating a foreground image indicating an area in which the object is present in the one image imaged by the one camera; comparing the depth image and the foreground image to determine whether there is a deficiency of a three-dimensional point in the three-dimensional model; and outputting deficiency information if it is determined that there is the deficiency of the three-dimensional point in the three-dimensional model.