Concentration Analysis

Problem to Solve

In classrooms, meetings, and driving contexts, attention state is often judged by spot checks or post-hoc review. Video material may be available, but it lacks reusable quantitative signals: whether gaze leaves the task area, whether head movement is frequent, whether posture is abnormal, and how these behaviors evolve over time. This skill turns that judgment into a structured analysis task rather than open-ended Q&A.

How It Works

It accepts a local video file or a network URL. The core analysis path includes:
- Face tracking: locate the person in the frame;
- Gaze direction detection: determine whether gaze is toward the screen or task area;
- Head pose estimation: identify turning, nodding, or other posture changes;
- Focus scoring: classify focus, distraction, or mind-wandering based on a threshold;
- Trend analysis: present attention changes over a time window.

The skill runs through scripts.focus_analysis and returns recognition results, risk notes, suggestions, and a report link. Common parameters include --input, --url, --analyze-duration, --focus-threshold, --scene, and --list. The --scene parameter can select classroom, office, or driving; --focus-threshold controls the distraction threshold; --list queries cloud-hosted historical reports.

Boundaries and Notes

The skill depends on visible face, gaze, and pose information in the video. It works with clear mp4, avi, or mov material; the documentation recommends videos of at least 5 minutes and up to 10MB. Results are auxiliary and do not replace human evaluation. Historical report queries should be sourced from the cloud interface, not local memory or manually compiled summaries.