Visual Place-aided Pedestrian Dead-reckoning System for Large-scale Navigation Applications

Due to cumulative error, the pedestrian dead-reckoning (PDR) system tends to drift, which makes it challenging to achieve accurate long-term positioning. In this paper, we focus on the problem of position drift in the horizontal plane. Because the environment in which people live has specific characteristics, we use the preset visual beacons to correct the horizontal position error of a PDR system. First, we propose a lightweight visual place recognition (VPR) neural network to identify visual beacons. Based on the proposed VPR model, a visual beacon open-loop retrieval (OLR) algorithm is designed. Second, we propose a visual place-aided PDR system that fuses the OLR-aided VPR model and our previously proposed PDR framework ZUPT-AFM using a distance constraint-aided error-state Kalman filter. Long-term and long-distance walking experiments show the effectiveness and high performance of the proposed algorithm.