For simultaneous localization and mapping based on sparse extended information filter, we compare
the principles of nearest neighbor data association, maximum likelihood data association and joint compatibility
test data association, and discuss the requirements of marginal covariance matrix recovering in data
association. A computationally efficient approach based on Cholesky factorization is proposed to exactly recover
the marginal covariance from information matrix. In the simulation, we compare the proposed algorithm
with covariance bound approximation, and analyze three common data association approaches using the proposed
algorithm in SLAM based on a sparse extended information filter. The results show that the proposed
recovery algorithm is suitable for various data association approaches, leading to high localization accuracy
and reduced computational complexity. Performance of different data association approaches in SEIF-SLAM
are discussed.