Detecting Walking Challenges in Gait Patterns Using a Capacitive Sensor Floor and Recurrent Neural Networks.

Area of Science:

• Biomechanics and Human Locomotion
• Machine Learning in Healthcare
• Sensor Technology

Background:

• Gait patterns provide insights into human health and disease progression.
• Analyzing gait disturbances can reveal underlying health conditions or recovery status.
• Current methods for gait analysis may lack objective, real-time data capture.

Purpose of the Study:

• To describe an experimental setup using a capacitive sensor floor for capturing gait patterns.
• To develop and evaluate machine learning models for gait analysis.
• To explore the potential of sensor floor data for health and care applications.

Main Methods:

• A capacitive sensor floor was utilized to record foot contact time and position data.
• A dataset of 42 participants walking in various modes (normal, eyes closed, dual-task) was collected.
• Recurrent neural networks (Long Short-Term Memory) were trained for walking mode classification and Unilateral Heel-Rise Test prediction.

Main Results:

• The recurrent neural network achieved promising results in classifying walking modes based on sensor floor data.
• A separate neural network instance successfully predicted the number of repetitions in the Unilateral Heel-Rise Test.
• The system demonstrated the feasibility of using sensor floor data for gait-related health assessments.

Conclusions:

• The combination of a capacitive sensor floor and recurrent neural networks is a viable system for gait analysis.
• This approach shows potential for objective health monitoring and rehabilitation tracking.
• Further research is warranted to translate these findings into practical health and care solutions.