Comparison of feature detection and outlier removal strategies in a mono visual odometry algorithm for underwater navigation
Author : Alessandro Bucci
Year : 2020
- Underwater images requests high exposure time, low image aquisition rate => feature matching instead of feature tracking
- pre processing with CLAHE
VIO:
- 2 matches per feature then filtering with the Lowe criterion
- And the outlier rejection is done with modified ICP or RANSAC or RANSAC+ICP
- The scale ambiguity is removed with the altimeter
- 25 feature matches is the minimum for VIO to work
- Estimate the transformation
$T_{k/k-1}, T_{k/k-2}, T_{k/k-3}$ to add additional constraints on the pose graph - UKF-based estimator with amixed kinematic-dynamic vehicle model, where only the longitudinal dynamics is taken into account (more details in An unscented Kalman filter based navigation algorithm for autonomous underwater vehicles)
Conclusion:
- ICP run time way higher than RANSAC and less precise