The detection quality of EVA is determined by clearly setting the Scenario, Detection Target, Vision Model, and Area and verifying them with actual alerts.

<aside> 🎯

What to judge is set to the scenario and frame mode, What to look for is set to the detection target and vision model, and Where to look is set to the detection area.

</aside>

Accurate detection starts from the connection in the settings

When detection is not as accurate as desired, rather than just modifying the scenario text, you should first think about which elements are problematic. Finding the target on the screen is the role of the vision model and detection target, and determining whether the target is an actual alarm situation is the role of the scenario and Agent. Where and how often the judgment is made is determined by the area and detection interval. Understanding this connection can help reduce false and false positives much more systematically.

When setting up for the first time, set the notification criteria one by one in the recommended order below and check the actual monitoring results. Then, as you build up field results from alert confirmations and feedback, and only if the problem recurs, you can make adjustments one by one, referring to detection quality improvements and meta-intelligence recommendations, to clearly see the effect of the changes.

Documentation Guide

Check the detection settings using the flow below.

  1. On this page, configure the scenario, detection target/model, area, and detection interval.
  2. Review the notifications and judgments that occurred in Confirm Notifications.
  3. Record actual results with true positive/false positive feedback.
  4. Adjust the scenarios and settings under Improve detection quality.
  5. Review improvements suggested by Meta-Intelligence Recommendations.

Recommended setup order

  1. Define detection situations, exceptions and frame modes as scenarios.
  2. Set the detection target and detection sensitivity.
  3. Select a vision model that matches the detection target characteristics.
  4. Specify the detection area to exclude unnecessary background.
  5. Set the detection interval.
  6. Start monitoring and calibrate with notifications and feedback.

Situation judgment