This guide explains how to reduce false positives and missed detections by adjusting the scenario, objects, Vision model, detection area, and detection interval based on actual alerts and feedback. First identify the issue in Alert Review and True-Positive and False-Positive Feedback.

Before Making Changes

Review recent alerts, detection images, and true-positive or false-positive feedback before changing a setting. Check whether the issue repeats and whether it occurs only in specific conditions such as a certain time, lighting, or camera angle.

Change one setting at a time so you can compare results under similar conditions.

Update Scenario Details

When objects are detected correctly but the situation decision is wrong, revise the scenario details to match site conditions.

1. Review the recent alert image and the Agent decision.
2. Rewrite the required alert condition in a short, specific sentence.
3. Add a specific exception when a recurring normal situation causes false positives.
4. Consider Sequence or Timeline when the decision requires a flow of scenes or duration.
5. Check alerts under similar conditions after the change and leave feedback.

For example, if glass reflections at an entrance are judged as flames, write a condition such as: Exclude reflections in the entrance glass and notify me only when actual smoke or flames appear.

<aside> 💡

Use conditions that can be confirmed in the image. Conditions requiring unseen intent, emotion, or assumptions can reduce decision accuracy.

</aside>

Checks by Problem Type

Objects Are Missed

1. Confirm that the object name is specific and visible in the image.
2. Check whether an object box appears in the stream.
3. Lower object-detection sensitivity slightly and compare results.
4. Confirm that the Vision model suits the object.