Related Experiment Video
Updated: Sep 7, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.1K
Inception-LSTM Human Motion Recognition with Channel Attention Mechanism
Yongtao Xu1,2, Liye Zhao1,2
1School of Instrument Science and Engineering, Southeast University, Nanjing 210096, China.
Computational and Mathematical Methods in Medicine
|June 23, 2022
Summary
This study introduces an Inception-LSTM algorithm using inertial sensors for human motion recognition, achieving high accuracy and offering an interpretable alternative to video-based methods.
Area of Science:
- Computer Science
- Biomedical Engineering
- Signal Processing
Background:
- Traditional video-based human motion recognition faces challenges like high cost, blind spots, and environmental sensitivity.
- Inertial sensor-based approaches offer a potential solution but require robust feature extraction and temporal modeling.
Purpose of the Study:
- To propose an improved Inception-LSTM algorithm for human motion recognition using inertial sensor signals.
- To enhance recognition accuracy and interpretability compared to existing methods.
Main Methods:
- Input inertial sensor signals are processed using an Inception structure for multi-scale spatial feature extraction.
- An improved Efficient Channel Attention (ECA) module refines critical details from spatial features.
- A Long Short-Term Memory (LSTM) network is employed for temporal feature extraction and motion classification.
Main Results:
- Achieved 95.04% recognition accuracy on the public PAMAP2 dataset.
- Attained 98.81% accuracy on a self-built dataset, demonstrating superior performance.
- Visual analysis of channel attention weights confirmed model interpretability and alignment with human intuition.
Conclusions:
- The proposed Inception-LSTM algorithm effectively recognizes human motion from inertial sensor data.
- The model offers a highly accurate and interpretable solution, overcoming limitations of video-based recognition.
