Information Filter

Parent: General Kalman Filter

Source: Scaradozzi 2018 SLAM application in surgery

• also same assumptions as the EKF
• main difference: how the Gaussian belief is represented
• est. cov. — replaced by information matrix (IM)
• est. state — replaced by information vector (IV)
• superior to KF in the following ways
  • data is filtered by summing up the IMs and IVs
  • often numerically more stable

Dual character of KF and IF

• in IF: prediction step requires two matrix inversions
  • higher computational complexity
  • high-dimension state space
  • while for KF, update in the prediction step is additive
• roles are reversed in measurement step

Variants of IF

• EIF (extended IF)
• SEIF (sparse extended IF)

Inspired by SLAM filters that are used to represent relative distances