Visual SLAM

Classic Approaches to Visual SLAM

Davison, ICCV 2003
*Traditional SLAM approach (Extended Kalman Filter)
*Maintains full camera and feature covariance
*Limited to Gaussian uncertainty only

Nistér, ICCV 2003
*Structure-from-motion approach (Preemptive RANSAC)
*Frame-to-frame motion only
*Drift: No repeatable localization

Recent Approaches to Visual SLAM

Pupilli & Calway, BMVC 2005
*Traditional SLAM approach (Particle Filter)
*Greater robustness: handles multi-modal cases
*New features not rigorously initialized

Eade & Drummond, 2006
*FastSLAM approach (Particle Filter/Kalman Filter)
*Particleper camera hypothesis, Kalman filter for features
*Allows larger maps: update O(MlogN) instead of O(N2)