Home

Welcome to the LF-VIO wiki! Our wiki is constantly updated......

Camera Model:

For large FoV cameras, there are already many models. The MEI camera model and the Scaramuzza camera model are currently supported in our program.

Scaramuzza camera model calibration tool: OCamCalib

MEI camera model calibration tool：Omnidirectional Calibration Toolbox

...

Dataset:

Indoors datasets are open sourced as follows:

ID01, ID06, ID10: Google Drive

ID01~ID10: Baidu Yun, Code: d7wq

IDL01~IDL02: Baidu Yun, Code: khw2

Initialization for Large FoV camera

• Epipolar geometry problem

Problem formulation:

$\mathbf{x}_2^T\left( \left[\mathbf{t}^{O_2}_{O_1}\right]^\wedge \mathbf{R}^{O_2}_{O_1} \right)\mathbf{x}_1=0$

$\mathbf{x}_1=\begin{bmatrix} \alpha_1 \\\beta_1 \\\gamma_1\end{bmatrix},\mathbf{x}_2=\begin{bmatrix} \alpha_2 \\\beta_2 \\\gamma_2\end{bmatrix}$

Code:

Eigen::Matrix3d InitialEXRotation::compute_E_21(vector<Eigen::Vector3d> &bearings_1, vector<Eigen::Vector3d> &bearings_2)

Input: bearings_1 and bearings_2 Output: best_E

• Triangulation problem
• Perspective-n-Point(PnP) problem ......

Optimization after initialization for Large FoV camera

• Visual measurement residual
• Inertial Measurement Unit(IMU) residual
• Marginalized residual