3D Shape Estimation via Multi-View Silhouette Projection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for reconstructing a three-dimensional shape of an object from multi-viewpoint images, such as the volume intersection method, require significant computational resources, especially when dealing with large spaces or high-resolution images, leading to increased calculation burdens.
Innovation Solution
The proposed solution involves an image processing apparatus that acquires silhouette image data from multiple viewpoints, generates low-resolution data, and calculates intersection intervals of line segments with the object by projecting them onto both low-resolution and high-resolution silhouette data, thereby reducing the computational load by limiting the processing to necessary regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the volume intersection method is used to reconstruct three-dimensional shape from multi-viewpoint images, then the shape estimation accuracy is improved, but the calculation amount increases significantly when the space to be measured is large
Solution Approach 1:
The patent segments the measurement space into a coarse grid structure where only certain regions (those intersected by rays from silhouette boundaries) are processed in detail. This divides the large space into manageable segments, reducing the overall calculation amount while maintaining shape estimation accuracy in the regions of interest.
Solution Approach 2:
The patent applies different processing qualities to different regions: high-resolution processing is applied only to regions where rays intersect silhouette boundaries (potential object regions), while other regions use lower-resolution or pre-defined grid data. This local differentiation reduces calculation amount without compromising the accuracy of the actual object shape estimation.
2Quantity of substance
If the contour of silhouette is extracted to limit ray intersection calculation, then the calculation amount is reduced, but the calculation amount increases if the resolution of input image is high
Solution Approach 1:
The patent extracts only the essential silhouette boundary information from high-resolution images, rather than processing the entire high-resolution image data. By taking out just the contour information needed for ray intersection testing, the calculation amount is reduced while maintaining the precision needed for accurate shape estimation.
Solution Approach 2:
The patent uses a hybrid approach where silhouette extraction is performed at reduced resolution for most rays, but full-resolution silhouette data is used for critical rays that define object boundaries. This partial application of high-resolution processing reduces overall calculation while maintaining necessary precision.
Data Source
AI summary
An image processing apparatus includes: an acquisition unit configured to acquire pieces of silhouette image data of an object viewed from multiple different viewpoints; a generation unit configured to generate, from the pieces of silhouette image data, pieces of low-resolution data representing images with a resolution lower than the pieces of silhouette image data; and an estimation unit configured to, by performing, for a plurality of line segments in space containing the object, processing in which, after a line segment in the space is projected onto a piece of low-resolution data to calculate a first intersection of the line segment with the object, the line segment is projected onto a piece of silhouette image data to calculate a second intersection of the line segment with the object, calculate intervals over which the plurality of line segments intersect the object and estimate a shape of the object.