Implementation of a real-time fall detection system based on hybrid threshold analysis algorithm and machine learning...

Area of Science:

• Gerontology
• Biomedical Engineering
• Computer Science

Background:

• The aging global population necessitates advanced health monitoring solutions.
• Accidental falls pose significant health risks to the elderly, demanding effective detection systems.
• Timely post-fall information delivery is crucial for elderly care.

Purpose of the Study:

• To develop and evaluate a hybrid fall detection system for the elderly.
• To improve the accuracy and efficiency of fall detection compared to existing methods.
• To create a system suitable for real-world application with lower power consumption and faster alerts.

Main Methods:

• Utilizing acceleration and angular velocity time series to capture human motion features.
• Implementing a hybrid algorithm combining edge-based threshold analysis and cloud-based machine learning.
• Classifying between falls and activities of daily living (ADLs) using the hybrid approach.

Main Results:

• The hybrid fall detection system achieved 98.55% accuracy, 98.16% sensitivity, and 98.73% specificity.
• The hybrid algorithm demonstrated lower power consumption and shorter average alarm times compared to a single machine learning algorithm.
• Performance was superior to single-threshold algorithms and comparable to solely cloud-based machine learning.

Conclusions:

• The proposed hybrid fall detection system offers a highly accurate and efficient solution for elderly fall monitoring.
• Edge-cloud computation in the hybrid model optimizes power consumption and response time for practical deployment.
• This system is well-suited for real-world applications, enhancing safety for the aging population.