Impact Detection in Fall Events: Leveraging Spatio-temporal Graph Convolutional Networks and Recurrent Neural...

Area of Science:

• Gerontology
• Biomedical Engineering
• Computer Science

Background:

• Falls are a major cause of accidental death in individuals over 65, posing a global health challenge.
• Existing fall detection systems struggle with accurately identifying impacts within fall events.
• Distinguishing actual impacts from non-impact falls is crucial for effective intervention and resource allocation.

Purpose of the Study:

• To propose an efficient and accurate methodology for detecting impacts during falls in elderly individuals.
• To enhance the precision of fall detection systems by differentiating between false alarms and genuine impact events.
• To improve healthcare resource allocation through more accurate fall impact identification.

Main Methods:

• Utilized 3D joint skeleton data represented as a graph.
• Employed spatio-temporal graph convolutional networks (STGCNs).
• Integrated gated recurrent unit (GRU) and bidirectional long short-term memory (BiLSTM) layers for impact detection.

Main Results:

• Achieved accuracy exceeding 90% in detecting impacts across various fall scenarios.
• Demonstrated the effectiveness of the proposed STGCN, GRU, and BiLSTM methodology.
• Successfully distinguished between false falls and actual impacts.

Conclusions:

• The developed methodology offers a significant advancement in accurate fall impact detection for the elderly population.
• This approach can lead to more precise healthcare responses and better management of fall-related incidents.
• The publicly released UP-Fall dataset will support future research in fall detection technology.