Spline-Based Initialization of Monocular Visual–Inertial State Estimators at High Altitude

Author: Liu

Year: 2017

Notes:

poor init conditions with poor parallax or excessive integration period of IMU
Ensures enough excitation of IMU to recover metric scale
Vision and IMU have different criterias for proper init:
- vision needs high parallax, not too dynamic motion
- IMU needs agile motions (not constant velocity)
Pb especially present at high altitude as high parallax needs long and constant velocity travel
A B-spline is fitted on a up-to-scale VO trajectory to exctract derivatives of the trajectory, then the acceleration is aligned with the spline second derivative
The method is also tested to bootstrap EKF and MSCKF
Rotational components is substracted from images using EIS (electronic image stabilisation)
states to be init: scale, gravity aligned attitude and velocity
6 order B spline to fit the traj
write a joint problem formulation of both B-spline fitting and states alignement
gyro bias are estimated in a stationnary position, they do not appear in the formulation
Observability analysis of $b_a$ :
- See the evolution of residual with perturbations on $\delta b_a$
- Only with enough rotation bias and gravity can be destinguished
- The bias are not incorporated in init