EL-SLE: Efficient Learning Based Stride-Length Estimation Using a Smartphone.

Area of Science:

• Computer Science
• Biomedical Engineering
• Human-Computer Interaction

Background:

• Pedestrian stride length estimation is vital for health monitoring and indoor localization apps.
• Current algorithms struggle with complex gaits and varied devices, leading to significant inaccuracies.
• There is a need for robust and adaptable stride length estimation models.

Purpose of the Study:

• To propose an efficient, learning-based stride length estimation (SLE) model using smartphones.
• To enhance the accuracy and generalization of SLE across heterogeneous devices, gaits, and environments.
• To develop an adaptive learning model that can update and improve its performance over time.

Main Methods:

• Utilized a smartphone-based approach for stride length estimation.
• Employed adaptive learning with Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) modules.
• Implemented a direct fusion method to map combined features to stride length.
• Integrated an online learning module for continuous model updates and generalization.

Main Results:

• The proposed model demonstrated superior performance compared to existing state-of-the-art methods.
• Achieved an average stride length estimation error rate of 4.26% across diverse experimental conditions.
• Validated effectiveness with heterogeneous devices, varied gaits, and dynamic environmental scenarios.

Conclusions:

• The proposed learning-based SLE model offers a significant advancement in accuracy and adaptability.
• Smartphone-based stride length estimation is feasible and can be highly accurate with advanced machine learning.
• The model's online learning capability ensures sustained performance and generalization in real-world applications.