Efficient Simultaneous Localization and Mapping Based on Ceiling-View: Ceiling Boundary Feature Map Approach

Abstract

For simultaneous localization and mapping (SLAM) based on the extended Kalman filter, the size of the state vector is an essential factor because the feasibility depends on it. In this paper, a new SLAM based on ceiling vision (cv-SLAM) is proposed. To keep the size of the state vector compact, the boundaries between ceiling and walls are used as landmarks for visual SLAM (vSLAM). The ceiling boundaries are robust to illuminative variations and they are not as numerous as the point features. Some constraints are imposed on the features based on the characteristics of the ceiling boundaries and an efficient update method called 'double update' is proposed to improve the SLAM performance. The basic idea of the double update is to fully utilize the intersections of the boundary features. Finally, the proposed SLAM is applied to some simulations and experiment, and its effectiveness is demonstrated through them.