Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Deep Learning-based Real-time Smartphone Pose Detection for Ultra-wideband Tagless Gate

Junyoung Choi, Sagnik Bhattacharya

TL;DR

Pose-induced localization errors can degrade UWB-based proximity gates. D3 fuses an LOS/NLOS classifier trained on CIR data with an IMU-driven pose detector on Android, employing an efficient 135-CIR representation (eCIR) to enable real-time operation. The approach achieves LOS/NLOS accuracy of 0.984 and pose accuracy of 0.961, while markedly reducing CIR transfer latency to 17.8 ms in a real-world 8×8 m test area. This work enables robust, pose-aware UTG operation and advances practical deployment of UWB-based proximity services.

Abstract

As commercial interest in proximity services increased, the development of various wireless localization techniques was promoted. In line with this trend, Ultra-wideband (UWB) is emerging as a promising solution that can realize proximity services thanks to centimeter-level localization accuracy. In addition, since the actual location of the mobile device (MD) on the human body, called pose, affects the localization accuracy, poses are also important to provide accurate proximity services, especially for the UWB tagless gate (UTG). In this paper, a real-time pose detector, termed D3, is proposed to estimate the pose of MD when users pass through UTG. D3 is based on line-of-sight (LOS) and non-LOS (NLOS) classification using UWB channel impulse response and utilizes the inertial measurement unit embedded in the smartphone to estimate the pose. D3 is implemented on Samsung Galaxy Note20 Ultra (i.e., SMN986B) and Qorvo UWB board to show the feasibility and applicability. D3 achieved an LOS/NLOS classification accuracy of 0.984, and ultimately detected four different poses of MD with an accuracy of 0.961 in real-time.

Deep Learning-based Real-time Smartphone Pose Detection for Ultra-wideband Tagless Gate

TL;DR

Pose-induced localization errors can degrade UWB-based proximity gates. D3 fuses an LOS/NLOS classifier trained on CIR data with an IMU-driven pose detector on Android, employing an efficient 135-CIR representation (eCIR) to enable real-time operation. The approach achieves LOS/NLOS accuracy of 0.984 and pose accuracy of 0.961, while markedly reducing CIR transfer latency to 17.8 ms in a real-world 8×8 m test area. This work enables robust, pose-aware UTG operation and advances practical deployment of UWB-based proximity services.

Abstract

As commercial interest in proximity services increased, the development of various wireless localization techniques was promoted. In line with this trend, Ultra-wideband (UWB) is emerging as a promising solution that can realize proximity services thanks to centimeter-level localization accuracy. In addition, since the actual location of the mobile device (MD) on the human body, called pose, affects the localization accuracy, poses are also important to provide accurate proximity services, especially for the UWB tagless gate (UTG). In this paper, a real-time pose detector, termed D3, is proposed to estimate the pose of MD when users pass through UTG. D3 is based on line-of-sight (LOS) and non-LOS (NLOS) classification using UWB channel impulse response and utilizes the inertial measurement unit embedded in the smartphone to estimate the pose. D3 is implemented on Samsung Galaxy Note20 Ultra (i.e., SMN986B) and Qorvo UWB board to show the feasibility and applicability. D3 achieved an LOS/NLOS classification accuracy of 0.984, and ultimately detected four different poses of MD with an accuracy of 0.961 in real-time.
Paper Structure (18 sections, 1 equation, 11 figures, 4 tables)

This paper contains 18 sections, 1 equation, 11 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Background and related work
  3. Background
  4. Related work
  5. Preliminary: UWB data collection
  6. Channel impulse response and received signal quality
  7. Definition of four poses of the mobile device
  8. CIR and IMU data collection
  9. $D^3$: Proposed system
  10. Overview
  11. Effective channel impulse response
  12. LOS/NLOS classifier
  13. Pose detector
  14. Performance Evaluation
  15. Implementation
  16. ...and 3 more sections

Figures (11)

  • Figure 1: The normal and unexpected operations of UTG. The opening and closing of the gate are colored in green and red, respectively.
  • Figure 2: The procedure of DS-TWR with ranging messages.
  • Figure 3: The example of CIR data including channel diagnostics. The values of FP_INDEX and maxNoise are 747 and 1409, respectively.
  • Figure 4: The illustration of four different poses for LOS and NLOS.
  • Figure 5: The diagnostics/CIR and IMU data collection during DS-TWR.
  • ...and 6 more figures