Classification of Indoor Human Fall Events Using Deep Learning.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Gerontology

Background:

• Falls are a frequent and serious concern for the elderly, necessitating advanced detection systems.
• Existing sensor-based alarm systems require improved accuracy and reliability for fall event classification.
• Distinguishing falls from natural indoor activities is crucial for effective monitoring.

Purpose of the Study:

• To propose a novel architecture for classifying human fall events from other indoor activities.
• To develop a reliable system for fall detection to assist in emergency response and healthcare.
• To enhance the capabilities of sensor-based alarm systems for elderly care.

Main Methods:

• Utilized a video frame generator to extract sequential data from video clips.
• Employed a two-dimensional convolutional neural network (2DCNN) for spatial feature extraction from video frames.
• Integrated a gated recurrent unit (GRU) network to capture temporal dependencies in human movement.
• Implemented a binary cross-entropy loss function for network training and optimization.
• Applied a sigmoid classifier for binary classification of fall events.

Main Results:

• The proposed model achieved a high accuracy of 99% in detecting human fall events.
• The system demonstrated superior performance compared to existing state-of-the-art models.
• Effective classification of fall events from natural indoor human activities was demonstrated.

Conclusions:

• The developed architecture offers a highly accurate solution for human fall identification.
• This technology can significantly contribute to the development of advanced sensor-based alarm systems for fall prevention and response.
• The model's performance highlights its potential for real-world application in elderly care and safety monitoring.