Elderly Long-Term Immobility Monitoring (>12h)

Problem Context

Solo-living elders who suffer a stroke, fall, or syncope may be unable to get up and call for help. Conventional one-off anomaly detection often focuses on explicit fall events and may miss a more hidden risk: sustained absence of visible activity. This skill targets solo-elder households and community elderly-care scenarios. Its goal is to generate an emergency alert when no human activity is detected for a configured window, such as 12h, so family members, grid workers, or care agencies can visit and verify.

How It Works and Limits

The skill analyzes video streams from fixed cameras in multiple zones. It accepts local files or video URLs and produces a structured report. Core capabilities include:

• Cross-zone activity detection: identifies whole-body movement, limb motion, hand gestures, and posture changes.
• Inactivity tracking: records the last activity timestamp, cumulative no-activity duration, and zone coverage.
• Threshold-based alerting: outputs levels such as none, warning, critical, and emergency based on a default 12h or configurable limit.
• Historical report lookup: queries cloud-hosted reports and renders them as a Markdown table.

Operators should note that inputs are generally limited to mp4/avi/mov files up to 10MB; night-vision and all-day footage are better suited for this analysis. The skill only performs visual activity monitoring; it does not provide health diagnosis or rescue procedures. When an emergency alert is triggered, immediate phone follow-up or in-person verification is required. For sensitive areas such as bathrooms, radar or PIR sensors should be preferred over cameras, and informed consent from the elder or family is required.