Skip to main navigation Skip to search Skip to main content

A Smartphone-Based Platform for Real-Time Deep Learning Pedestrian Dead Reckoning

  • Xuehang Sun
  • , Yan Wang
  • , You Li
  • , Zhihao Liu

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Pedestrian Dead Reckoning (PDR) is crucial for indoor positioning when GNSS is not available. However, on-device deployment and unified testing platforms for deep learning-based PDR are still scarce. As a highly integrated and widely adopted terminal, the smartphone not only incorporates multimodal sensors such as IMU, camera, and GNSS, but also leverages a on-device computing resources capable of executing deep models locally. Thus, smartphones are a good choice for the development and testing of PDR. Based on this insight, this paper proposes a model conversion and lightweight deployment scheme and develops a deep learning PDR testing and visualization platform that can be installed on smartphones. The platform adopts a modular architecture that supports real-time execution of mainstream models and provides online visualization and logging of metrics such as global 2D trajectories, local 3D trajectories, and inference latency. Experiments show that on-device real-time inference is consistent with desktop offline inference in trajectory reconstruction and error evaluation. The proposed platform significantly reduces deployment and testing costs, promoting the practical implementation of deep learning-based PDR. © 2025 IEEE.
Original languageEnglish
Title of host publication2025 International Ubiquitous Positioning, Indoor Navigation and Location-Based Services Conference, UPINLBS 2025
PublisherInstitute of Electrical and Electronics Engineers Inc.
ISBN (Print)979-833157138-2
DOIs
StatePublished - 17 Dec 2025

Publication series

NameInt. Ubiquitous Position., Indoor Navig. Locat.-Based Serv. Conf., UPINLBS

Indexed by

  • Scopus

Fingerprint

Dive into the research topics of 'A Smartphone-Based Platform for Real-Time Deep Learning Pedestrian Dead Reckoning'. Together they form a unique fingerprint.

Cite this