Dynamic Initialization Problem

[QingfengLi-hit]

The current initialization scheme requires static conditions. If the static conditions are not good, it will drift away later. I plan to use SFM to solve , but with this method, the covariance in the sliding window cannot be obtained, and only the result of the state(ba, bg, pose,v) can be used. Is there any good initialization scheme for msckf?

[goldbattle]

Here are some papers that might get you started, hope this helps.

Closed-form solution of visual-inertial structure from motion Agostino Martinelli https://hal.archives-ouvertes.fr/hal-00905881/document

Visual-Inertial Monocular SLAM with Map Reuse Raul Mur-Artal, Juan D. Tardos https://arxiv.org/abs/1610.05949

Robust initialization of monocular visual-inertial estimation on aerial robots Tong Qin; Shaojie Shen https://ieeexplore.ieee.org/document/8206284

Revisiting visual-inertial structure from motion for odometry and SLAM initialization Georgios Evangelidis, Branislav Micusik https://arxiv.org/abs/2006.06017

An Analytical Solution to the IMU Initialization Problem for Visual-Inertial Systems David Zuñiga-Noël, Francisco-Angel Moreno, Javier Gonzalez-Jimenez https://arxiv.org/abs/2103.03389