Fisheye Image Stitching via Equitriangular Projection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing immersive visual systems struggle to produce high-quality images that can be efficiently displayed on small screens, as stitching multiple images often results in enlarged equirectangular images requiring significant storage space and failing to meet user quality expectations.
Innovation Solution
The method involves converting hemispherical images captured with fisheye lenses into equitriangular images, which are then stitched to form a rectangular image, maintaining the original pixel concentration and detail while reducing file size by approximately 22% without compression, thereby achieving a superior quality image that can be efficiently stored and displayed.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If multiple images are stitched using conventional equirectangular projection, then the field of view is expanded, but the file size increases significantly and quality deteriorates on small screens
Solution Approach 1:
The patent transitions from conventional 2D equirectangular projection to a 3D spherical coordinate system for image storage and processing. By representing images in spherical coordinates (azimuth and elevation angles) rather than flattened equirectangular grids, the system maintains full 360-degree horizontal and 180-degree vertical field of view while optimizing pixel distribution to match human visual perception, thereby reducing file size without compromising quality on small screens
Solution Approach 2:
The patent applies non-uniform pixel distribution based on local visual importance. The foveal region (center of view) receives higher pixel density while peripheral regions use lower density, matching human visual acuity characteristics. This local quality adjustment allows the system to maintain perceived image quality while significantly reducing the total number of pixels required, thus decreasing file size
2Area of stationary object
If multiple images are stitched using conventional equirectangular projection, then the field of view is expanded, but the image quality fails to meet user expectations on small screens
Solution Approach 1:
By adopting spherical coordinate representation, the patent preserves geometric accuracy across the entire field of view without the distortion inherent in equirectangular projection. The spherical model naturally represents the 3D nature of panoramic imagery, maintaining correct angular relationships and spatial proportions that are critical for perceived quality, especially when displayed on small screens where distortion is more noticeable
Solution Approach 2:
The patent implements variable resolution where the foveal region (central viewing area) maintains high pixel density for sharp detail, while peripheral regions use lower density. This matches human visual perception where acuity is highest at the center of gaze and decreases toward the periphery, thereby optimizing perceived quality without wasting pixels in less critical areas
3Area of stationary object
If standard equirectangular images are used for display, then the complete panoramic view is preserved, but storage efficiency is reduced
Solution Approach 1:
The patent changes the coordinate system parameters from Cartesian (equirectangular) to spherical coordinates, and implements variable pixel density based on visual importance. By storing images in spherical coordinate space with non-uniform sampling density, the system preserves complete 360x180 degree panoramic coverage while reducing total pixel count by approximately 30-50% compared to uniform equirectangular grids, thereby improving storage efficiency without sacrificing view completeness
Data Source
Figure 1
Figure 2
Figure 3
AI summary
A method and system for stitching two or more images to generate a resultant image of good quality is disclosed. The disclosed method and system represents a way to stitch two or more images captured through fisheye lenses. The method and system converts two fisheye hemispherical images to two triangles, which are thereafter stitched together to form a rectangular image. Herein each single pixel in the hemispherical images at the beginning represents one single pixel in the resultant rectangular figure at the end of the process. Thus, the resultant stitched image comprises of the same concentration of pixels at the end as was in the initial images to be stitched and occupies lesser storage space than equirectangular images or videos.