Improved Sparsing of Extended Information Filter for Vision-SLAM

Abstract

There are a number of algorithms used on Simultaneous Localization and Mapping(SLAM). The sparse extended information filter (SEIF) algorithm is deduced by the sparsification treatment to EIF algorithm, which is the information form of EKF. SEIF has been successfully implemented with a variety of challenging real-world data sets and has lead to new insights into scalable SLAM. However, the computational burden related to information matrix balloons with respect to the increase of the mapped landmarks, the most computational cost is in recovery information matrix (inverse matrix).In this paper, by analyzing the every steps of information matrix update in SEIF process, A sparsification rule is put forward, which enforce the elements into zero according to setting a threshold and computer inverse information matrix with the method of tri-diagonal matrix splitting according to observation information of sparsification time. the computational complexity is much lower by using new sparsification rule. The algorithm used in vision-SLAM shows that the computational complexity of the SEIF algorithm is a constant, which is independent of environment features. That means SEIF has a high value of application in large-scale environment with a large number of features.