Spherically Optimized RANSAC Aided by an IMU for Fisheye Image Matching

Author: Liang

Year: 2021

Notes:

image matching method: use fisheye spherical coordinate and provide relative rotation with IMU to get pose estimation and mismatch removal.

Related work:

references of image matching for fisheye using rectified image (but lose of FOV, and image artifacts)
Spherical SIFT, omni SIFT and also affine invariant detector (hessian affine)
many sensor fusion methods
many deep learning based methods

System:

Use the polynomial fisheye model of openCV
Use the spherical model for estimating Essential matrix
hypothesis: when IMU and camera are fixed rotation obtained by the camera is the one of the IMU (no calibration needed)
four point method with RANSAC (one parameter being fixed by IMU)

Experiments:

Evaluation with soa algorithm on the matching success rate
Discussion about IMU noise by adding noise on transformations obtained with MARG => very interesting

Commentaire:

Article modeste mais trés intéressant