Abstract:The large field of view (FOV) of fisheye cameras makes them widely applied in many fields. However, a fisheye image contains bigger distortions than a traditional image does, resulting in traditional algorithms of the feature point-based image matching cannot simply be transplanted to a fisheye image. For this reason, an algorithm for feature point pseudo-neighborhood extraction is proposed. This idea is based on many ray beams around a feature point having the same incident FOVs when imaging with a fisheye camera. Thus, the model starts from incident ray beams of FOV around a given feature point. The algorithm first establishes the initial FOV-curves on the projection surface of the fisheye hemisphere, and then constructs an extended 2-D curve, called "FOV-curve network." Finally, the feature point pseudo-neighborhood is constructed using the FOV-curve network. The four datasets are used to validate the proposed algorithm. The experimental results demonstrate that the average improvement rate of the matching point pairs for three datasets in outdoor scenes reached 23.5%, 319.6%, and 86.6% in the center zone, the left margin, and the right margin zone, respectively. The biggest one is for Dataset-2, which reaches 27.8% and 714% in the center and left margin zone when compared with three types of traditional speeded-up robust feature (SURF), FAST, and ORB algorithms. An average improvement rate relative to the deep learning, SuperGlue model, for three datasets in outdoor scenes are 2.8%. These experimental results demonstrate that the proposed algorithm effectively eliminates the impact of fisheye image distortion on feature point-based matching.

Achieving widely distributed feature matches using flattened-affine-SIFT algorithm for fisheye images

FMS:Features Motion Statistics for Incorrect Matched-pair Removal

Linear fitting interpolation based on FOV division for correcting wide angle fish-eye lens distortion

Extraction of Feature Point Pseudo-Neighborhood for High-Accuracy of Fisheye Image Matching

Real-time and High Precision Feature Matching Between Blur Aerial Images

Fish-eye image correction method and technique based on geometrical imaging model

Calibration of Fisheye Cameras Based on the Viewing Sphere

Unified Optical Distortion Correction Method for Imaging Systems Using A Concise Geometrical Transformation Model

Fisheye image Distortion Correction Based on Spherical Perspective Projection Constraint

Two distortion correcting methods for fisheye images

A rectification algorithm for un-calibrated multi-view images based on SIFT features

A Comprehensive Overview of Fish-Eye Camera Distortion Correction Methods

Fisheye lens distortion correction

A Fast Correction Algorithm for Inverse Longitude and Latitude Mapping of Fish Eye Image

Distortion correction of fisheye lenses using a nonparametric imaging model

A computationally efficient affine-invariant feature for image matching with wide viewing angles

A Stronger Stitching Algorithm for Fisheye Images based on Deblurring and Registration

Correction of Barrel Distortion in Fisheye Lens Images Using Image-Based Estimation of Distortion Parameters

Efficient Covisibility-based Image Matching for Large-Scale SfM

A High Precision Feature Matching Method Based On Geometrical Outlier Removal For Remote Sensing Image Registration

Geometry-aware Feature Matching for Large-Scale Structure from Motion