Skip to main content

Detection Scenario

The first step to begin AI analysis in the EVA system is to create a Detection Scenario. A detection scenario is more than just identifying the presence of an object — it defines rules that allow AI to trigger alerts only under specific desired conditions, based on its understanding of the scene.



Refining the Scenario​

The more precisely a detection scenario is written, the better the EVA Agent can understand the situation. To reduce false alarms, it is especially important to clearly describe situations where alerts should be excluded or specify exception conditions.

By continuously reviewing the alerts that occur and refining the detection scenario, EVA can make increasingly accurate judgments and provide reliable results without unnecessary alerts.



Writing Scenarios with the Generate Feature​

EVA provides a Generate feature to help users easily create detection scenarios.

  1. Enter a Draft: Briefly describe the situation you want to detect.

  2. Click Generate: EVA Agent’s LLM analyzes the input and automatically generates a scenario template that includes Detection Steps and Exception Conditions.

  3. Review and Refine: By reviewing and manually refining or adjusting the generated template, you can achieve far more accurate detection results.

💡 TIP: The scenario suggested by EVA is a draft. By checking the actual alert results and gradually adjusting the detection conditions, EVA will increasingly make optimal judgments tailored to your environment.



Example of a Generated Detection Scenario​

Example – Detecting a Person Who Has Fallen

User Query: "Notify me if a person has fallen."

Target:

  • A person lying on the floor

Detection Steps:

  • A person is present
  • At least one person appears to have fallen

Exceptions:

  • The fallen person’s body is partially hidden (e.g., occluded)
  • The person’s full shape is unclear (only silhouette or partial view, such as shoes)
  • The fallen person appears unharmed (e.g., looking at a phone, lying on a desk)
  • The image quality is too low to make an accurate judgment


Automatic Vision ML Target Extraction​

EVA automatically extracts Vision ML detection targets from the scenario description provided by the user. This helps clearly define the detection objective and improve the recognition performance of the Vision model.

Example Input: "Notify me if there is a fire or even a small amount of smoke." Target: [Fire, Smoke]



Situation Analysis and Alert Evaluation​

Based on the written detection scenario, when specific conditions are detected, the EVA Agent analyzes the appropriateness of the situation using the VLM along with image snapshots.

Depending on this analysis, the system decides whether to trigger an alert, and users receive detailed messages describing the detected event.

Continuously review the alert results if unnecessary alerts occur, try refining the detection conditions step by step.

Through this iterative improvement, EVA will perform more precise alert judgments tailored to your specific operational goals.