Impact Detection in Fall Events: Leveraging Spatio-Temporal Graph Convolutional Networks and Recurrent Neural Networks Using 3D Skeleton Data

Researchers proposed a 3D-skeleton-based fall-impact detection model using spatio-temporal graph convolutional networks with GRU and BiLSTM layers.

The method analyzes 3D joint data to identify the moment of impact during a fall, improving on standard fall detection.

On an improved UP-Fall dataset, the model achieved over 90% accuracy.

Better impact detection could reduce false alarms and improve monitoring and response for older adults.