An Approach for Fall Prediction Based on Kinematics of Body Key Points Using LSTM.

Area of Science:

• Gerontology
• Computer Science
• Biomedical Engineering

Background:

• Existing fall prediction methods often rely on wearable sensors or video analysis of body posture and kinematics.
• Integrating these approaches can enhance the accuracy and reliability of fall detection systems.

Purpose of the Study:

• To develop and evaluate machine learning models for accurate fall prediction in adults and the elderly.
• To leverage body kinematics extracted from video data for enhanced fall detection capabilities.

Main Methods:

• Utilized the UP-Fall Detection dataset, comprising video recordings.
• Developed three Long-Short-Term Memory (LSTM) network models (4p-SAFE, 5p-SAFE, 6p-SAFE) using coordinates and angles extracted from video sequences.
• Employed machine learning for analyzing sequential image data from ordinary cameras.

Main Results:

• Achieved a prediction accuracy of up to 98% with the developed LSTM models.
• Demonstrated the effectiveness of using body kinematics and machine learning for fall prediction.
• The models are designed for easy application with standard camera footage.

Conclusions:

• The proposed integrated approach of body kinematics and machine learning offers a highly accurate method for fall prediction.
• Implementation of these models can significantly improve the health and wellness of adults and the elderly.
• The system's reliance on ordinary cameras makes it widely applicable, particularly in aged-care settings.