Author: Sun
Year: 2017
- Multi-State Constraint Kalman Filter (MSKF) to reduce computationnal cost
- Demonstrates that stereo is more robust for a similar computatinonal cost w.r.t. moncular solution
- represents state with standard aditive error so that
$\delta q = [\theta 1]^T$ - visual measurements with stereo observations
- project the residual on the null space of the jacobian wrt the lmk pose (classic)
- 4 unobservable DoF in Kalman Filter VIO (global position and yaw)
- explanations in Observability-based Rules for Designing Consistent EKF SLAM Estimators
- uses KLT w/ FAST keypoints for both temporal and stereo tracks: * Empirically, corner features with depths greater than 1m can be reliably matched across the stereo images using KLT tracking with a 20cm baseline stereo configuration*