Deep Learning Model for Robust Target Tracking Using TDoA Probabilistic Image

Sungho Lee, Jaewoong Shim

Research output: Contribution to journalArticlepeer-review

Abstract

Ultra Wide Band (UWB) based indoor positioning methods utilizing Time Difference of Arrival (TDoA) are commonly used; however, their performance significantly degrades in environments with high Additive White Gaussian Noise (AWGN). Although many studies have attempted to remove AWGN from TDoA, these approaches require additional positioning methods and lack robustness in various environments. In this paper, we propose a 'TDoA Probabilistic Image Based Target Tracking (TPITT)' method that robustly estimates an object's position using TDoA with AWGN. TPITT generates probability images of object presence in each region using TDoA and employs a 'Convolution-LSTM' model to estimate object coordinates. Experiments demonstrate the proposed method's robustness and low prediction error across diverse environments. Notably, TPITT is more effective than the prior study 'TDoA Image Based Target Tracking (TITT)' in environments with high AWGN.

Original languageEnglish
Pages (from-to)807-815
Number of pages9
JournalJournal of Korean Institute of Communications and Information Sciences
Volume48
Issue number7
DOIs
StatePublished - Jul 2023

Keywords

  • Deep learning
  • Indoor tracking
  • Real time localization
  • TDoA
  • UWB communication systems

Fingerprint

Dive into the research topics of 'Deep Learning Model for Robust Target Tracking Using TDoA Probabilistic Image'. Together they form a unique fingerprint.

Cite this