A tridentnet-lstm pedestrian dead reckoning method based on prior physical model constraint

By processing shoulder IMU data using the TridentNet-LSTM method, and combining prior physical models and deep learning, the accuracy and robustness issues of pedestrian dead reckoning in shoulder-wearing scenarios are solved, achieving high-precision trajectory reconstruction in complex environments.

CN122170850APending Publication Date: 2026-06-09SOUTHEAST UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
SOUTHEAST UNIV
Filing Date
2026-03-05
Publication Date
2026-06-09

AI Technical Summary

Technical Problem

In complex environments, shoulder-mounted pedestrian dead reckoning methods face challenges such as high sensor noise and irregular dynamic patterns, resulting in insufficient robustness and large trajectory accumulation errors in existing methods, and difficulty in handling feature ambiguity issues in shoulder IMU data.

Method used

The TridentNet-LSTM method based on prior physical model constraints is adopted, which combines feature extraction, multi-scale feature fusion and temporal modeling. The shoulder IMU data is processed through an end-to-end deep learning model, and smoothness and velocity constraints are introduced to improve the accuracy and consistency of the trajectory.

Benefits of technology

It effectively processes shoulder IMU data and is suitable for walking, running, and walking-running gaits, improving the accuracy and robustness of pedestrian dead reckoning and ensuring that the trajectory conforms to the laws of physical motion.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122170850A_ABST
    Figure CN122170850A_ABST
Patent Text Reader

Abstract

This invention discloses a TridentNet-LSTM pedestrian dead reckoning method based on prior physical model constraints, comprising: constructing a TridentNet-LSTM model to combine multi-scale feature extraction with time series modeling; constructing a loss function consisting of three parts: a basic loss function based on weighted mean square error loss, a sequence consistency loss function, and a Physical Information Neural Network (PINN) loss function; during model training, data preprocessing is first performed, and the training and validation sets are divided and initialized; the Adam optimizer is selected for end-to-end training, and the ReduceLROnPlateau learning rate scheduling strategy is adopted; the PDR algorithm is used for trajectory reconstruction, and RMSE is used as the error analysis metric. This invention, through the TridentNet-LSTM model based on prior physical model constraints, can achieve end-to-end pedestrian trajectory reckoning and can adapt to walking, running, and mixed walking-running gaits. It does not rely on the strong periodic signals of foot IMUs and can also handle shoulder IMU data with relatively ambiguous gait characteristics.
Need to check novelty before this filing date? Find Prior Art