Direct visual orientation estimation of omnidirectional images
Based on [LibPeR] C++
library, this repository contains multiple examples of code to demonstrate various use cases of direct visual orientation estimation.
Summary of available methods and omnidirectional image types:
Photometric | Mixture of Photometric Potential | |
---|---|---|
Dual fisheye | folder | folder |
Equirectangular | folder | folder |
OmniCV 2022 CVPR Workshop [paper]:
André, A. N., & Caron, G. (2022). Photometric Visual Gyroscope for Full-View Spherical Camera. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 5232-5235).
Based on dual fisheye omnidirectional images, compute the relative orientation by minimizing the photometric cost.
Example available in folder P_SSD_gyroEstimation/
OmniCV 2023 CVPR Workshop [paper]:
Berenguel-Baeta, B., André, A. N., Caron, G., Bermudez-Cameo, J., & Guerrero, J. J. (2023). Visual Gyroscope: Combination of Deep Learning Features and Direct Alignment for Panoramic Stabilization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 6444-6447).
Based on equirectangular omnidirectional images, compute the relative orientation by minimizing the photometric cost.
Example available in folder P_SSD_gyroEstimation_EquiRect/
ICRA 2018 [paper]
Caron, G., & Morbidi, F. (2018, May). Spherical visual gyroscope for autonomous robots using the mixture of photometric potentials. In 2018 IEEE International Conference on Robotics and Automation (ICRA) (pp. 820-827). IEEE.
Based on dual fisheye omnidirectional images, considers each vertex of the spherical mesh as a mixture of photometric potentials and minimizes the overall cost to retrieve the orientation.
Example available in folder MPP_SSD_gyroEstimation/
OmniCV 2023 CVPR Workshop [paper]:
Berenguel-Baeta, B., André, A. N., Caron, G., Bermudez-Cameo, J., & Guerrero, J. J. (2023). Visual Gyroscope: Combination of Deep Learning Features and Direct Alignment for Panoramic Stabilization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 6444-6447).
Based on equirectangular omnidirectional images, considers each vertex of the spherical mesh as a mixture of photometric potentials and minimizes the overall cost to retrieve the orientation.
Example available in folder MPP_SSD_gyroEstimation_EquiRect/
OmniCV 2023 CVPR Workshop [paper]:
Berenguel-Baeta, B., André, A. N., Caron, G., Bermudez-Cameo, J., & Guerrero, J. J. (2023). Visual Gyroscope: Combination of Deep Learning Features and Direct Alignment for Panoramic Stabilization. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (pp. 6444-6447).
Install the following dependencies:
- LibPeR that contains the methods for visual orientation estimation of omnidirectional images (tested with 0.5.0)
- ViSP (tested with 3.5)
- Boost (tested with 1.71.0)
- LibXML2
Compile the example:
mkdir build && cd build
cmake ..
make -j12
Run the compiled program
To perform the visual orientation estimation, the examples rely on some common parameters. For an exhaustive list, please read the explanations attached to each example.
- Path to omnidirectional images. [link] to PanoraMIS dataset used in the articles or shorter versions just to test with dual-fisheye images or equirectangular images
- Number of subdivision of the icosahedron: any value in {3,4,5}
-
iRef
reference image number -
i0
first image to measure the orientation -
i360
last image to measure the orientation -
iStep
step used to process the images -
Mask
if needed to occult unwanted parts of the images to align -
nbTries
if$>0$ , perform initial guesses on an estimated orientation, before estimating the orientation using the optimization law -
estimationType
- 0 for pure gyroscope
- 1 for odometry
- 2 for odometry with key images
-
stabilization
1 to compensate and stabilize the output computed images or 0 to simply save output unstabilized images
Some parameters are specific to the image used or to the method:
lambdaG
Sets the Gaussian expansion forMPP
-based methodscalibFile
provides the intrinsic parameters of the dual fisheye camera, as a.xml
file
Copyright (C) 2017-2024 by MIS lab (UPJV). All rights reserved.
See https://mis.u-picardie.fr/~g-caron/fr/index.php?page=7 for more information.
This software was developed at:
MIS - UPJV
33 rue Saint-Leu
80039 AMIENS CEDEX
France
and at
CNRS - AIST JRL (Joint Robotics Laboratory)
1-1-1 Umezono, Tsukuba, Ibaraki
Japan
This file is provided AS IS with NO WARRANTY OF ANY KIND, INCLUDING THE
WARRANTY OF DESIGN, MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE.
Description:
Insight about how to set the project and build the program
Authors:
Guillaume CARON, Antoine ANDRE