Related Experiment Video
Updated: Nov 7, 2025
06:37
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
4.7K
A Lightweight Attention-Based CNN Model for Efficient Gait Recognition with Wearable IMU Sensors
Haohua Huang1,2, Pan Zhou1,2, Ye Li1
1Joint Engineering Research Center for Health Big Data Intelligent Analysis Technology, Shenzhen Institute of Advanced Technology, Chinese Academy of Sciences, Shenzhen 518055, China.
Sensors (Basel, Switzerland)
|April 30, 2021
Summary
This study introduces a lightweight deep learning model for wearable gait recognition using attention-based Convolutional Neural Networks (CNNs). The model achieves high accuracy while significantly reducing complexity for practical use on wearable devices.
Area of Science:
- Computer Science
- Biomedical Engineering
- Signal Processing
Background:
- Wearable sensors enable gait recognition for identity verification.
- Deep learning models have advanced gait recognition accuracy.
- Existing models often lack the efficiency required for wearable devices due to high complexity.
Purpose of the Study:
- To develop a lightweight, attention-based Convolutional Neural Network (CNN) model for efficient wearable gait recognition.
- To enhance feature extraction and reduce model complexity for practical application on resource-constrained wearable devices.
Main Methods:
- A four-layer lightweight CNN was utilized for initial gait feature extraction.
- A novel attention module incorporating contextual encoding and depthwise separable convolution was integrated to refine features and reduce model size.
- Softmax classification was employed for the final gait recognition task.
Main Results:
- The proposed model demonstrated high recognition performance on the whuGait and OU-ISIR datasets.
- Comprehensive experiments analyzed the impact of attention mechanisms and data segmentation on performance.
- The model achieved an average reduction in complexity of 86.5% compared to existing methods while maintaining superior accuracy.
Conclusions:
- The developed lightweight attention-based CNN model offers a promising solution for accurate and efficient wearable gait recognition.
- The proposed attention module effectively enhances gait features and significantly simplifies model complexity.
- This approach is suitable for deployment on wearable devices, balancing performance with computational efficiency.
