Fall detection from a manual wheelchair: preliminary findings based on accelerometers using machine learning...

Area of Science:

• Biomedical Engineering
• Machine Learning
• Assistive Technology

Background:

• Automated fall detection systems are crucial for individuals using wheelchairs to mitigate fall-related injuries.
• Existing fall detection technologies often do not adequately address the unique mobility patterns and fall risks associated with wheelchair use.

Purpose of the Study:

• To develop and train a machine learning algorithm for accurate fall detection in wheelchair users.
• To differentiate between actual falls and common wheelchair mobility activities using sensor data.

Main Methods:

• Utilized machine learning techniques, specifically Neural Network classifiers, to train a fall detection algorithm.
• Collected data from accelerometers placed on the wrist, chest, and head of participants simulating falls and performing wheelchair activities.
• Trained the algorithm on data from 258 simulated falls and 220 wheelchair mobility activities.

Main Results:

• The developed algorithm demonstrated excellent accuracy in differentiating falls from wheelchair mobility.
• Sensors achieved high accuracy: wrist (100%), chest (96.9%), and head (94.8%).
• This pilot study confirms the algorithm's effectiveness in a controlled laboratory setting.

Conclusions:

• A machine learning-based fall detection algorithm using accelerometer patterns can effectively distinguish wheelchair falls from mobility activities.
• Integration into wrist-worn devices for real-world testing among community-based wheelchair users is recommended.