Child Hazardous Behavior Recognition Tool

Problem to Solve

Children at home, in kindergartens, or nurseries may engage in risky actions such as climbing, playing with fire, touching electrical sources, or lingering near windows. Manual monitoring can miss incidents, and generic video summaries do not provide a structured judgment about hazardous behavior. This skill targets surveillance video and turns the result into a reviewable answer: whether hazardous behavior occurred, how often, whether it crossed the alert threshold, and where the report is located.

How It Works

• Input: local video files or network video URLs; supported formats are mp4/avi/mov, up to 10MB.
• Analysis: the skill calls the API through scripts/child_dangerous_behavior_recognition_analysis.py and evaluates human skeletal keypoints and behavioral trajectories in the video stream.
• Alerting: --alert-threshold sets the warning score, defaulting to 0.6; values above the threshold produce an alert.
• Output: it returns a structured report with risk/recognition results, recommendations, and a report link; history listings are queried from the cloud API via --list and rendered as a Markdown table.

Boundaries and Notes

This skill is an auxiliary safety-monitoring capability, not a replacement for immediate guardian confirmation after an alert. Internal identity handling is performed automatically by the script; users should not be asked to provide or exposed to identity values. Historical reports must be read from the cloud interface rather than local memory or manually summarized records.