Introduction
The AI Settings section controls how the AI camera detects people, analyzes faces, measures attention, and collects audience analytics.
If AI Triggers determine what content should be displayed, then AI Settings determine how the system identifies and analyzes the audience standing in front of the screen.
Simply put:
AI Settings = How the camera sees and understands people.
AI Triggers = What content is shown to those people.
These settings allow administrators to balance:
- Audience reach
- Data quality
- Recognition accuracy
- Attention tracking quality
- Analytics reliability
- System performance
1. Camera
The Camera section controls the basic behavior of the camera, including image orientation, video quality, and processing performance.
Camera Rotation (0° / 90° / 180° / 270°)
This setting rotates the camera image. It is used when the camera is physically installed in a different orientation.
Examples:
- 0°: standard installation
- 90°: camera mounted sideways
- 180°: camera mounted upside down
- 270°: camera mounted sideways in the opposite direction
Correct camera rotation is important because all AI analytics depend on the camera image, including: Face detection, Audience tracking, Attention tracking, ROI calculations, Trigger activation.
If the rotation is incorrect, the AI may analyze the audience improperly.
FPS Limit
FPS (Frames Per Second) determines how many video frames the AI processes every second. Think of it as the smoothness of the video stream and how frequently the AI receives new information.
Higher FPS: Smoother tracking, Faster reactions, More processing power required.
Lower FPS: Reduced system load, Lower CPU usage, Slightly slower detection response.
Examples: 10–15 FPS is sufficient for most audience analytics applications. 30 FPS provides smoother tracking and faster updates. Higher values increase hardware requirements.
Recommended: For most installations, 10–15 FPS provides the best balance between performance and accuracy.
Resolution Preset
This setting determines the camera image quality. Available options may include: 720p, 1080p.
Higher resolutions provide: Better face visibility, Improved demographic analysis, Better age and gender estimation, Improved recognition at greater distances.
However, higher resolutions also: Increase processing requirements, Require more bandwidth, May reduce overall performance on lower-powered devices.
Recommended: 720p is ideal for most installations. 1080p is recommended when faces need to be analyzed from a greater distance and sufficient processing power is available.
2. Attention
The Attention section controls how the system measures viewer attention. The goal is not simply to detect that a person exists, but to determine whether that person is actually looking at the screen.
Attention Tracking
This setting enables or disables attention measurement.
Enabled: The AI attempts to determine whether a person is actively looking at the display.
Disabled: The AI can still count audience presence, but attention metrics will not be collected.
Attention Sensitivity
This setting determines how long someone must look at the screen before attention is recorded.
Example: If set to 3 seconds, a person must maintain attention for approximately three seconds before the system counts it as engagement.
Why is this useful? It prevents accidental glances from being counted as genuine attention.
Examples: A person walks by and glances for one second → not counted. A person stops and watches for five seconds → counted as attention.
Recommended Values: 1–2 seconds → aggressive tracking. 3 seconds → balanced default setting. 5+ seconds → strict attention validation.
3. Audience Filters
The Audience Filters section controls who is considered a valid audience member and how strict the AI should be when analyzing people.
Sensitivity
Sensitivity controls how strict the AI should be when analyzing demographic information. This affects: Age estimation, Gender detection, Emotion detection, Demographic confidence.
Important: Sensitivity affects demographic analysis quality. It does not control person identification or Hash ID generation. Those are controlled by the Confidence Threshold setting.
Low Sensitivity: Designed to maximize audience reach. Ideal for: Shopping malls, Hallways, High-traffic environments, Walk-by installations. The AI accepts faces even when: Faces are partially visible, People are moving, Lighting conditions are not ideal, Faces appear briefly. Advantages: More audience detected, Fewer missed visitors. Trade-Off: Demographic accuracy may be lower.
Medium Sensitivity: Balanced mode. Provides a compromise between: Audience volume, Analytics quality, Detection stability. This is the recommended default setting for most deployments.
High Sensitivity: Designed for maximum demographic accuracy. Ideal for: Advertising campaigns, Analytics-focused projects, Digital signage measurement, Premium reporting environments. The AI requires clearer facial visibility before demographic analysis is performed. Advantages: More accurate age estimation, More accurate gender classification, Better emotion analysis. Trade-Off: Fewer people may qualify for demographic analysis.
Stability Time
This setting defines how long a person must remain visible before being considered a valid audience member.
Example: If Stability Time is set to 3 seconds, a person must remain in view for at least three seconds before being counted.
Why is this useful? It prevents random passersby from affecting analytics.
Recommended Values: 1 second → fast-moving environments. 3 seconds → standard configuration. 5–10 seconds → focus only on highly engaged viewers.
Audience Re-Detection Delay
This setting determines how long the system waits before treating a returning visitor as a new audience session.
Example: If set to 60 seconds: Person leaves and returns after 20 seconds → same session. Person leaves and returns after 2 minutes → new session.
Why is this useful? It prevents the same person from being counted repeatedly when briefly stepping out of view.
4. Detection
The Detection section controls face detection quality, recognition reliability, and simultaneous face processing.
Confidence Threshold
Confidence Threshold determines how certain the AI must be before accepting a face detection or recognition result. This affects: Face Detection, Face Recognition, Hash ID generation, Re-identification accuracy, Visitor matching.
Important: Confidence Threshold affects recognition quality. It does not control age, gender, or emotion accuracy. Those are controlled by Sensitivity.
Lower Confidence Values (Example: 50%): The AI is more permissive. Advantages: More faces detected, Faster recognition, Higher audience volume. Trade-Off: Increased chance of duplicates, Lower recognition consistency.
Higher Confidence Values (Example: 90%): The AI requires stronger confidence before creating or matching a visitor profile. Advantages: Better recognition consistency, Fewer duplicate profiles, More reliable Hash IDs. Trade-Off: Some faces may not qualify, Detection becomes stricter.
Recommended: Start with 50–70% and increase only if stronger recognition quality is needed.
Min Face Size
This setting defines the minimum face size required for AI analysis. In simple terms: If a face is too small, the system ignores it.
Lower Values: Detects faces farther away, Captures more audience, Higher processing load.
Higher Values: Only larger faces are analyzed, Better quality data, Distant people may be ignored.
Recommended: Use lower or medium values if audience members are expected to stand farther from the screen.
Maximum Number of Faces Processed Simultaneously
This setting limits how many faces the AI can analyze at the same time.
Example: If set to 6, the AI will process up to six faces simultaneously. If ten people appear, only a portion may be analyzed.
Why is this useful? It controls system load and resource usage.
Higher Values: More people analyzed, Increased CPU usage.
Lower Values: Lower system load, Some faces may be skipped.
Recommended: 4–6 faces is sufficient for most deployments.
5. Live Preview / ROI
This section provides a live view of the camera and allows administrators to define the active analysis area.
Live Preview: Displays exactly what the camera currently sees. It helps verify: Camera placement, Camera angle, Audience visibility, AI operation in real time. Administrators can quickly confirm that people appear within the intended viewing area.
Current FPS: Displays the actual frame rate currently being processed. If FPS becomes too low, camera performance may appear unstable. For smooth operation, maintaining 10–15 FPS or higher is recommended.
Faces Detected: Displays the number of faces currently recognized by the AI. This is useful for verifying that face detection is working correctly. If a person is visible but the counter remains at zero, the camera placement or AI settings may need adjustment.
Run Test: The test function performs a short AI validation session. A typical test helps verify: Face detection quality, Camera stability, Audience visibility, AI responsiveness. This is especially useful after changing AI settings.
6. ROI (Region of Interest)
ROI defines the area of the camera image that should be used for AI analysis. In simple terms: ROI tells the AI where it should pay attention. Anything outside the ROI is ignored.
Full Screen ROI: In Full Screen mode, the AI analyzes the entire camera view. This is the fastest and simplest configuration. Recommended when: The camera only sees the intended audience area, There are no unwanted background zones, The entire image is relevant. Limitation: The AI may analyze people who are not actually interacting with the display.
Polygon ROI: Polygon ROI allows administrators to manually draw a custom analysis area. Only people located inside the selected region will be analyzed. Recommended when the camera view includes: Walkways, Entrances, Employee areas, Cash registers, Irrelevant traffic zones. Example: The camera sees an entire room, but the display is located only on one side. A polygon can be drawn around the display viewing area so the AI ignores unrelated activity.
How ROI Works
People Outside ROI: The system should ignore them. They should not: Generate analytics, Create Hash IDs, Trigger content, Contribute to attention metrics.
People Inside ROI: The system can: Count audience presence, Analyze demographics, Measure attention, Activate AI triggers, Generate analytics.
Recommended AI Setup Workflow
- Verify the camera using Live Preview.
- Adjust Camera Rotation if needed.
- Select Resolution and FPS settings.
- Configure ROI to focus on the desired audience area.
- Enable Attention Tracking if engagement measurement is required.
- Select an appropriate Sensitivity level.
- Configure Stability Time to filter accidental passersby.
- Adjust Confidence Threshold for recognition quality.
- Run a system test.
- Save the configuration.
Simple Rule of Thumb
If you want to capture more audience: Use more permissive settings: Low Sensitivity, Lower Stability Time, Lower Confidence Threshold.
If you want higher quality analytics: Use stricter settings: High Sensitivity, Higher Stability Time, Higher Confidence Threshold. Polygon ROI.
Summary
AI Settings determine how accurately, efficiently, and consistently the system analyzes its audience. These settings directly affect: Audience counts, Demographic analytics, Hash ID quality, Attention tracking, AI Trigger performance, Overall system efficiency. A properly configured AI Settings profile ensures more accurate analytics, better audience understanding, and more reliable AI Trigger behavior.
