Enhanced and Facilitated Indoor Positioning by Visible-Light GraphSLAM Technique
Yuan Yue, Xiaohui Zhao, Zan Li
Abstract
Recently, indoor positioning has played a critical role in many emerging indoor applications. However, due to complicated indoor environments, it is still challenging to develop an indoor positioning system with high positioning accuracy and low deployment efforts. In this work, an indoor positioning system based on visible light fingerprinting is proposed by leveraging a novel visible light GraphSLAM (VL-GraphSLAM) technique. The proposed VL-GraphSLAM provides enhanced solutions at both frontend and backend to improve the accuracy of estimated trajectories. Then, the estimated trajectory is anchored in floor map based on a novel door detection method to recover indoor walking paths. Based on VL-GraphSLAM, we construct a database with the visible light received signal strength labeled by the locations of walking paths, which is called visible light map. Moreover, a Kalman filter is adopted to fuse the visible light fingerprinting and inertial sensors to locate users. Comprehensive experiments illustrate that our proposed system can accurately recover walking paths (0.4 m) and locate users (0.9 m) in an accuracy of submeter, which significantly outperforms a traditional WiFi-based fingerprinting system and is more convenient to deploy than a traditional visible light positioning based on ranging.