Video Behavior Monitor

idle
What this actually detects. Falling, staggering, tremor/shivering, drowsiness, unresponsiveness and grimacing come from real-time pose/face landmark tracking (heuristic, not clinical-grade). Yelling / screaming / loud talking and "calling for help" come from live audio analysis and speech keyword spotting.

What this does not detect. Vomiting, coughing, sneezing, urinating and defecating are not reliably identifiable from pose landmarks or simple video heuristics — they need an action-recognition model trained on labeled clips of those specific behaviors. See the classifyUnsupportedActions() stub in the script for where to plug in a real video-classification backend (e.g. periodically posting short clips to your own trained model / cloud video-understanding API). Treating this list as "handled" without that backend will just produce false alarms.

Live signals

Event log