Multi-Camera Image Stitching With Curved Horizon Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques for combining images from multiple cameras on a mobile body often result in composite images that appear bent at the horizon, due to the use of virtual projection surfaces with hemispherical and flat components.
Innovation Solution
An image processing system that acquires images from multiple external cameras mounted on a host mobile body, converts these images into viewpoint images viewed from a specific virtual viewpoint, and generates a composite image by synthesizing these viewpoint images, resulting in an imaged ground surface with a convex horizon and aligned positions representing the same actual ground surface location.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If images are combined using virtual projection surfaces with hemispherical and flat components, then the imaging coverage is improved, but the composite image appears bent at the horizon
Solution Approach 1:
The patent applies a curved projection surface with a convex horizon instead of traditional hemispherical and flat components. This curvature approach allows the ground surface to curve upward naturally, eliminating image bending artifacts while maintaining comprehensive imaging coverage across the field of view.
2Area of stationary object
If multiple camera images are synthesized to create a composite image, then the field of view is expanded, but parallax effects cause misalignment in the overlapped regions
Solution Approach 1:
The patent transforms images from multiple cameras by converting them to a common curved projection surface coordinate system. This parameter transformation approach adjusts the projection geometry to eliminate parallax-induced misalignment, ensuring that corresponding positions in overlapped regions accurately represent the same ground surface locations.
3Device complexity
If a flat projection surface is used for image synthesis, then the processing complexity is reduced, but the horizon appears straight and unrealistic
Solution Approach 1:
The patent employs a curved projection surface that naturally produces a convex horizon, creating a more realistic and visually accurate composite image. The curvature is mathematically defined to match the expected ground surface geometry, providing both aesthetic realism and geometric accuracy without significantly increasing processing complexity.
Data Source
AI summary
An image processing system includes a processor configured to acquire a first image by a first external camera and a second image by a second external camera, convert the first image and the second image into a first viewpoint image and a second viewpoint image from a specific virtual viewpoint, and generate a composite image by synthesizing the first viewpoint image and the second viewpoint image. The composite image has an imaged ground surface that curves to have a horizon convex upward. A first position on an imaged ground surface of the first viewpoint image and a second position on an imaged ground surface of the second viewpoint image are aligned with each other, and the first position and the second position represent the same position in an actual ground surface. An object extends across the horizon in the composite image has a smoothly curved outline across the horizon.