Outdoor Walking Classification Based on Inertial Measurement Unit and Foot Pressure Sensor Data.

Area of Science:

• Biomechanics and Human Movement Analysis
• Wearable Sensor Technology
• Machine Learning in Healthcare

Background:

• Gait patterns are significantly altered by different walking surfaces.
• Automatic classification of walking conditions is crucial for gait analysis and rehabilitation.
• Current methods require comparison of various sensor modalities and processing techniques.

Purpose of the Study:

• To develop and compare tools for automatic walking condition classification.
• To evaluate the effectiveness of different sensor modalities (IMUs, pressure insoles) and their combinations.
• To assess the impact of gait cycle segmentation versus sliding window approaches on classification performance.

Main Methods:

• Twenty participants performed walking trials on various surfaces (flat, stairs, slopes) while wearing IMUs and pressure insoles.
• Machine learning (Extreme Gradient Boosting) and deep learning (CNN+LSTM) models were trained for classification.
• Sensor modalities included lower-limb IMUs, foot IMUs, pelvis IMUs, pressure insoles, and combinations thereof.

Main Results:

• A deep learning model using lower-limb IMUs with gait segmentation achieved the highest performance (F1=0.89).
• IMU-based models significantly outperformed pressure insole models (p<0.01).
• The best performing minimal model combined pelvis IMUs and pressure insoles using a sliding window (F1=0.83).

Conclusions:

• Inertial measurement units (IMUs) offer the most discriminative features for classifying walking conditions.
• Deep learning models demonstrate strong performance without the need for gait segmentation.
• Combining sensor modalities can enhance classification accuracy, particularly for machine learning models.