Area-Aware 3D Feature-Point Reconstruction from Multiple Views
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for reconstructing three-dimensional coordinates of feature points from multiple viewpoints suffer from reduced accuracy due to the bilateral symmetry of human faces, particularly when images are captured obliquely, leading to inaccuracies in detecting feature points on one side of the face.
Innovation Solution
An image processing apparatus that detects feature points from multiple images, appends attribute information indicating the area of the object, and determines three-dimensional coordinates using a combination of viewpoints to minimize calculation errors, employing threshold values and error calculations to select optimal viewpoints.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If oblique viewpoint images are used for three-dimensional coordinate reconstruction, then the coverage of object areas is improved, but the measurement precision of feature points on the far side deteriorates
Solution Approach 1:
The patent segments the object into multiple areas (first area and second area) and assigns different viewpoint selection criteria to each area. For the first area, viewpoints where the area occupies a large image area are selected, while for the second area, viewpoints with smaller occlusion from other parts are selected. This segmentation approach allows each area to be reconstructed using optimal viewpoints specific to its characteristics.
Solution Approach 2:
The patent applies different quality criteria locally to different areas of the object. Instead of using a uniform viewpoint selection criterion for the entire object, it tailors the selection criteria to each area's specific characteristics - using area occupancy for one region and occlusion minimization for another, thereby optimizing measurement precision locally for each area.
2Measurement precision
If frontal viewpoint images are used for feature point detection, then the detection accuracy of all feature points is improved, but the ability to obtain three-dimensional coordinates deteriorates due to insufficient disparity
Solution Approach 1:
The patent merges multiple images captured from different viewpoints to reconstruct three-dimensional coordinates. By combining information from both frontal and oblique viewpoint images, it achieves both accurate feature point detection (from frontal views) and sufficient disparity information (from oblique views) for complete 3D reconstruction of all object areas.
3Area of stationary object
If multiple viewpoints are used for three-dimensional coordinate reconstruction, then the coverage and completeness of reconstruction is improved, but the complexity of viewpoint selection and processing increases
Solution Approach 1:
The patent performs preliminary actions by calculating image area ratios and occlusion degrees for all candidate viewpoints before actual 3D reconstruction. These pre-calculated metrics are stored and used to automatically select appropriate viewpoints for each object area, simplifying the subsequent reconstruction process and reducing real-time computational complexity.
Solution Approach 2:
The patent changes parameters by using different evaluation criteria (image area ratio for one area, occlusion degree for another) to select viewpoints for different object areas. This parameter-based approach provides a systematic and automated method for viewpoint selection, reducing the complexity of manual viewpoint selection while ensuring comprehensive reconstruction coverage.
Data Source
AI summary
Resulting from a reduction in the accuracy of a feature point obtained from a viewpoint inclined with respect to the front direction of an object, the estimation accuracy of three-dimensional coordinates of the feature point is reduced. Consequently, from each of a plurality of images captured from a plurality of viewpoints, the feature point of the object is detected and attribute information indicating which area of the object the detected feature point belongs to is appended to the detected feature point. Then, for each of the same attribute information, the three-dimensional coordinates of the feature point are calculated by using two-dimensional coordinates of the feature point on the image corresponding to the viewpoints not more than the plurality of viewpoints and not less than two.