Visible Light-Based Position and Pose Tracking: A Bidirectional Recurrent Convolutional Learning Method to Address Environment Dynamics

Abstract

In this paper, we are interested in visible light-based position and orientation tracking (VLP) for mobile user devices (UDs) in dynamic environments. Conventional model-based VLP usually depends on a perfect signal propagation model (SPM) with fixed parameters, and hence their performance will be seriously decreased when environment varies over time, e.g., due to diffuse scattering or receiver optical filter gain fluctuation. To address this challenge, in this paper we propose a bidirectional recurrent convolutional neural network (Bi-RCNN)-based VLP algorithm. Our Bi-RCNN extracts time-domain correlation feature of measurement sample series, and simultaneously spatial-domain texture feature is captured via 3D convolutional-driven memory networks. In this way, spatial-time texture features are fully exploited, and thus mobile UD tracking performance is improved. Numerical experiment validates that our Bi-RCNN-based VLP outperforms existing VLP baselines, and the achieved localization error is around 1.5 cm in dynamic environments.