Design and analysis of a framework for real-time vision-based SLAM using Rao-Blackwellised particle filters

This paper addresses the problem of simultaneous localization and mapping (SLAM) using vision-based sensing. We present and analyse an implementation of a Rao- Blackwellised particle filter (RBPF) that uses stereo vision to localize a camera and 3D landmarks as the camera moves through an unknown environment. Our implementation is robust, can operate in real-time, and can operate without odometric or inertial measurements. Furthermore, our approach supports a 6-degree-of-freedom pose representation, vision-based ego-motion estimation, adaptive resampling, monocular operation, and a selection of odometry-based, observation-based, and mixture (combining local and global pose estimation) proposal distributions. This paper also examines the run-time behavior of efficiently designed RBPFs, providing an extensive empirical analysis of the memory and processing characteristics of RBPFs for vision-based SLAM. Finally, we present experimental results demonstrating the accuracy and efficiency of our approach.