Bidirectional recurrent convolutional learning-based visible light position and pose tracking against environment variation
炳朋 周, 光森 陈, 杰友 朱 · Scientia Sinica Informationis · 2022
In this paper, we focus on visible light-based position and orientation tracking (VLP) for user devices in dynamic environments. Conventional signal model-based VLP methods depend on a perfect signal propagation model (SPM) with fixed parameters, and hence their performance is reduced when the localization environment varies over time, e.g., because of diffuse scattering, reflections, and receiver gain fluctuations. To address this challenge, in this paper, we propose a bidirectional recurrent convolutional neural network (Bi-RCNN)-based VLP algorithm. The Bi-RCNN extracts the time-domain correlation feature of a measurement sample series through bidirectional recurrent subnetworks and simultaneously employs a 3D convolutional neural network to capture spatial-domain texture features. In this way, the spatial-time texture features are fully exploited, thus improving the mobile device tracking performance. Numerical experiments show that our Bi-RCNN-based VLP solution outperforms the existing VLP baselines, and the achieved localization error is around 1.5 cm in dynamic environments under an SNR of 20 dB.