Description
Fall detection is a binary or multi-class subset of human-activity-recognition whose target is the abrupt vertical motion plus subsequent stillness that characterises a fall. It draws disproportionate attention because it is the most clinically actionable WiFi-sensing task in elderly-care deployments. Practical systems combine a Doppler-spectrogram trigger with a stillness verifier to reject false positives from sit-down events.
When it's used
- Privacy-preserving in-home elderly monitoring
- Hospital-room continuous safety monitoring
- Robustness benchmark separating "fall" from "lie down" / "sit"
Limitations
- Class imbalance — falls are rare in real data
- High cost of false positives (alarm fatigue) and false negatives (missed event)
- Requires multi-second observation window, delaying response