Skip to main content

Blue Iris 6 – AI Tuning

Overview​

This page provides practical guidance for using AI confirmation in Blue Iris to improve event quality and reduce false alerts.

The focus is on disciplined setup: motion first, AI second, alerts third.

🎯 Objective​

  • Improve event quality
  • Reduce false alerts
  • Build a stable AI confirmation workflow
  • Avoid using AI to compensate for poor trigger design

🧠 Key Concepts​

  • AI confirms events — it does not replace motion tuning
  • Image quality directly affects AI accuracy
  • Confidence thresholds must be tuned based on results
  • Stable workflows outperform aggressive configurations

🛠️ Core Tuning Areas​

Motion Before AI​

  • Establish clean motion detection first
  • Reduce false triggers caused by scene noise
  • Confirm cameras produce reliable trigger events

Image Source​

  • Ensure AI analyzes a usable image
  • Poor framing, compression, or lighting reduces effectiveness
  • Evaluate whether main stream or sub-stream produces better results

Confidence Thresholds​

  • Start with moderate confidence settings
  • Review confirmed vs rejected events
  • Adjust thresholds only after observing real-world behavior

Object Classes​

  • Enable only required object types
  • Avoid unnecessary classifications that create noise
  • Align detection with actual security objectives

Alert Logic​

  • Do not enable aggressive notifications initially
  • Validate AI-confirmed events first
  • Then introduce alerts and external integrations

Performance​

  • AI increases system load
  • Monitor CPU, GPU, and system responsiveness
  • Scale gradually when enabling across multiple cameras

📊 Recommended Baseline​

  • Clean motion detection
  • Moderate confidence threshold
  • Limited object classes
  • Simple alert logic
  • Review actual results before expanding

🧠 Real-World Notes​

  • Most AI issues originate from motion or scene design
  • Proper framing often improves results more than tuning
  • Excess object classes increase review workload
  • Controlled rollout produces more stable long-term performance

⚠️ Common Mistakes​

  • Enabling AI before motion is stable
  • Analyzing poor-quality images
  • Using too many object classes
  • Sending alerts before validation
  • Changing multiple variables simultaneously

📊 Related Pages​

✅ Result​

A practical framework for implementing AI confirmation in Blue Iris without introducing unnecessary noise or instability.