Efficient Monocular SLAM using sparse information filters

Abstract

A new method for efficiently mapping three dimensional environments from a platform carrying a single calibrated camera, and simultaneously localizing the platform within this map is presented in this paper. This is the Monocular SLAM problem in robotics, which is equivalent to the problem of extracting Structure from Motion (SFM) in computer vision. A novel formulation of Monocular SLAM which exploits recent results from multi-view geometry to partition the feature location measurements extracted from images into providing estimates of environment representation and platform motion is developed. Proposed formulation allows rich geometric information from a large set of features extracted from images to be maximally incorporated during the estimation process, without a corresponding increase in the computational cost, resulting in more accurate estimates. A sparse Extended Information Filter (EIF) which fully exploits the sparse structure of the problem is used to generate camera pose and feature location estimates. Experimental results are provided to verify the algorithm.