An improved Rao-Blackwellized particle filter based-SLAM running on an OMAP embedded architecture

Abstract

A monocular SLAM system uses a single-camera trying to solve the problem of simultaneous localization and mapping. The FastSLAM2.0 employs a Rao-Blackwellized particle filter to estimate the robot pose based on a set of hypotheses that represent the different possible trajectories, while mapping a large number of landmarks. The most common problem related with such a system is the initialization of the landmarks. The monocular camera is a bearing only sensor and can not provide the depth of the observed feature. A lot of methods was developed for an efficient estimation of the depth of the landmarks. The unified inverse depth parametrization allows an efficient and undelayed initialization of landmarks. This work present a full monocular SLAM system based on the FastSLAM2.0 algorithm. The algorithm is tested on a real dataset, optimized and then implemented on a low cost embedded architecture.