Image Taking Position Calculation for Narrow-Angle Camera 3D Reconstruction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Structure from Motion (SfM) techniques often result in large-scale non-linear optimization problems with many local solutions, leading to inconsistent three-dimensional image reconstructions, especially when using cameras with narrow angles of view or mobile terminals, making stable operation difficult and increasing calculation time.
Innovation Solution
A processing apparatus and method that calculates image taking positions by selecting optimum candidates based on correspondence points, using a storage unit, calculation unit, selection unit, and determination unit to converge on local solutions and determine image inclusion in the selected set, allowing for efficient processing of images from general cameras with narrow angles of view.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If SfM techniques are used to calculate image taking positions from multiple two-dimensional images, then three-dimensional shape reconstruction is achieved, but inconsistent three-dimensional images are produced due to large-scale non-linear optimization problems with many local solutions
Solution Approach 1:
The patent divides the set of multiple images into a reference image group and a target image. By calculating image taking positions for the target image relative to the reference image group separately, the method segments the global optimization problem into smaller local problems, reducing the impact of local solutions and improving calculation reliability
Solution Approach 2:
The patent introduces correspondence points between images as intermediary elements to establish relationships between the reference image group and target image. These correspondence points serve as mediators to calculate image taking positions without directly solving the complex global optimization problem, thereby improving consistency
2Ease of operation
If images are taken using a general camera with a narrow angle of view, then the camera is more readily available and easier to use, but the calculation time increases and processing becomes less efficient
Solution Approach 1:
The patent performs preliminary selection of reference images from the reference image group that have high similarity to the target image. By pre-selecting the most suitable reference images before calculation, the method reduces the computational burden and accelerates processing speed while maintaining accuracy for narrow-angle cameras
Solution Approach 2:
The patent changes the parameter of image selection by using similarity metrics to choose reference images. This parameter change allows the system to adapt to narrow-angle camera characteristics and improve calculation efficiency without requiring special wide-angle equipment
3Measurement precision
If multiple correspondence points are detected in a wide range to improve image matching accuracy, then a special camera with a wide-angle lens or fish-eye lens is required, but this increases device complexity and cost
Solution Approach 1:
The patent shifts from spatial dimension (wide-angle field of view) to similarity dimension (image content comparison). By selecting reference images based on similarity metrics rather than relying on wide-angle cameras to capture more correspondence points spatially, the method achieves accurate matching using standard cameras
Solution Approach 2:
The patent creates a virtual reference image group by selecting and copying relevant features from multiple reference images based on similarity. This virtual construction allows the system to work effectively with narrow-angle cameras without requiring physical wide-angle or fish-eye lenses
Data Source
AI summary
A three-dimensional shape of a subject is acquired in a healthy and realistic calculation time, using two-dimensional images taken by a general camera having a narrow angle of view. The processing apparatus calculates an image taking position of each of a plurality of images, and includes: a storage unit for storing thereinto a selected image set including images whose image taking position has been calculated; a calculation unit calculates a plurality of image taking position candidates for an unselected image of the plurality of images; a selection unit selects an optimum image taking position having a highest degree of coincidence in correspondence points, from among the plurality of image taking position candidates for the unselected image; and a determination unit determines whether or not the unselected image is added to the selected image set.