Confidence Metrics

Confidence-related outputs combine detection and tracking signals.

Typical Metrics

• Detection confidence (detector quality signal)
• Assignment confidence (match quality signal)
• Position uncertainty (state covariance-derived signal)

Why They Matter

• Identify hard frames for active learning.
• Detect parameter regimes that overfit or underfit scene dynamics.
• Prioritize manual review where trajectory quality is weakest.

Integration Points

• Collected during tracking worker pipeline.
• Optionally persisted to CSV when enabled.
• Consumed by dataset generation/scoring workflows.