Viewpoint Path Stabilization for 3D Image Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Handheld camera images captured while moving around an object are subject to significant distortions due to vertical movement and incomplete capture paths, leading to noise and variation in image sequences.
Innovation Solution
A method involving a processor that projects three-dimensional points onto image locations, determines transformations to align images with a smoothed path, and generates new images using neural networks to create a multiview interactive digital media representation (MVIDMR) navigable in multiple dimensions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If images are captured by a handheld camera moving around an object, then the camera can capture the object from multiple viewpoints, but the images are subject to significant distortions due to vertical movement and incomplete capture paths
Solution Approach 1:
The patent introduces a virtual camera positioned on a smoothed closed-loop path as an intermediary to generate synthetic images. This virtual camera acts as a mediator between the noisy handheld camera data and the desired clean multi-view representation, projecting 3D points from the object model to virtual camera positions to create aligned, distortion-free images
Solution Approach 2:
The patent performs preliminary actions by first creating a 3D model of the object from captured images, then pre-computing a smoothed closed-loop path around the object, and finally generating synthetic images from virtual camera positions on this path before actual handheld capture is complete or to correct its deficiencies
2Ease of operation
If the camera path includes vertical movement and is not entirely closed, then the capture is more flexible and easier to perform, but the image sequence contains noise and variation
Solution Approach 1:
The patent creates a idealized copy of the capture process by generating synthetic images from a virtual camera following a perfect closed-loop path. This copy replaces the flawed actual capture sequence, preserving the multi-view information while eliminating the inconsistencies caused by vertical movement and incomplete paths
Solution Approach 2:
The patent transforms the camera trajectory parameters by computing a smoothed closed-loop path that eliminates vertical movement variations and ensures complete coverage. The virtual camera's position and orientation parameters are continuously adjusted along this smoothed path to maintain consistent image quality
3Manufacturing precision
If transformations are applied to align images with a smoothed path, then viewpoint alignment is improved, but the processing complexity increases
Solution Approach 1:
The patent replaces complex mechanical image alignment operations with computational projections. Instead of physically adjusting cameras or performing complex geometric transformations on captured images, the system uses 3D point projections from a digital model to virtual camera positions, achieving precise alignment through mathematical computation rather than mechanical adjustment
Data Source
AI summary
Three-dimensional points may be projected onto first locations in a first image of an object captured from a first position in three-dimensional space relative to the object and projected onto second locations a virtual camera position located at a second position in three-dimensional space relative to the object. First transformations linking the first and second locations may then be determined. Second transformations transforming first coordinates for the first image to second coordinates for the second image may be determined based on the first transformations. Based on these second transformations and on the first image, a second image of the object from the virtual camera position.