1
Return

Fisheye image stitching by RANSAC-R and MRM

delete2026-08-03
delete0
PRE
AI
Y
Yuan Jia
C
Cailin Wu
R
Rui Song *
B
Bin Song
DOI:10.1007/s00521-026-12284-9delete
deleteOriginal
deleteOriginal request for help
deleteShare
deleteSave
Abstract

Abstract

En 中文
Fisheye cameras have been widely used in the field of panoramic imaging. However, most image stitching research has focused on pinhole camera images, with only a few studies addressing fisheye images. Moreover, in these studies, the distortion of fisheye images is frequently corrected to transform them into planar images prior to carrying out the subsequent stitching process. This paper proposes a novel inlier filtering algorithm, RANSAC Based on Rotation (RANSAC-R). Based on the principle of light incidence, we convert traditional matched points into matched light rays and then screen these matched light rays, which directly performs precise inlier matching on equidistant fisheye images. In the experimental evaluation, we demonstrate that the proposed RANSAC-R algorithm significantly improves matching performance. Specifically, the recall rate is increased from approximately $$50\%$$ to $$100\%$$ , while maintaining $$100\%$$ matching precision. Furthermore, in experiments conducted with 140-degree and 180-degree fisheye cameras, the inlier rate is improved by approximately 2.5 times and 4 times, respectively. When capturing images, it poses a significant challenge to ensure the precise alignment of the camera’s optical centers, resulting in parallax that introduces artifacts into the stitched images. To eliminate artifacts, this paper presents a new stitching method, the multi-rotation motion projection method (MRM), which is capable of producing high-quality stitching results for multiple fisheye or pinhole images without introducing noticeable distortion. The experimental results indicate that both methods substantially enhance stitching accuracy, effectively reduce geometric distortion, and yield panoramas of practical quality.
Keywords:
Fisheye
Pinhole
Image stitching
Panorama
RANSAC-R
MRM

Journal

Neural Computing and Applications cover
Neural Computing and Applications
IF:
4.5
Papers:
729
Citations:
3.2W

Organization

S
School of Telecommunications Engineering
Scholars:
16
Papers: 8
Citations: 0
Cited Papers

Cited Papers

Citing Papers

Citing Papers