EL-OGISLAM: Escalator Landmark-based Occupancy Grid Inertial SLAM framework

The positioning accuracy of inertial pedestrian navigation system (IPNS) in indoor multi-story building scenarios is still unsatisfactory. In this paper, we propose an Escalator Landmark-based Occupancy Grid Inertial SLAM (EL-OGISLAM) framework, which mainly uses the pedestrian motion vectors and the associated escalator landmarks to constrain the positioning error of system. We obtain the pedestrian motion vectors from the traditional foot-mounted IMU-based PNS (Foot-PNS) and the system localization accuracy is maintained to a certain extent under the OGI-SLAM framework. Subsequently, the escalator landmarks are extracted, and the position, type, and heading angle of the escalator are considered as landmark descriptors to determine the landmark association relationship. After we use the associated landmarks to optimize the positions of all landmarks through full bundle adjustment, the trajectories between adjacent landmarks are affine transformed to achieve accurate simultaneous localization and mapping, and overall reduce the localization error. The experimental results show that the proposed system can achieve an average horizontal error of 0.74 m and an average height error of 0.11 m in the case of pedestrians walking and taking escalators.