Welcome to the second article in our three-part series. Previously, we covered the fundamental principles behind DISPL’s AI-driven audience analytics—why lighting, angles, and camera quality matter, and how the system detects and analyzes faces. Now, let’s explore real-world scenarios to show what good vs. bad camera setups look like, highlight typical mistakes, and provide practical tips to avoid those pitfalls.
1. Why Camera Placement Matters
In order to detect and analyze faces accurately, each camera must capture a clear, well-lit, frontal view of people’s faces. If your camera is poorly placed, obstructed, or set at extreme angles, the AI cannot generate reliable face descriptors—or in some cases, it may fail to detect faces entirely.
Below are the most common pitfalls we see in retail, kiosk, and other in-store environments, followed by straightforward recommendations to help you get the best results.
2. Illustrative Examples: Correct vs. Incorrect Placement
2.1 Correct Placement
- Height & Angle: Camera mounted at 1.5–2.5 meters high (roughly eye level), angled slightly downward or straight on, ensuring the face is within ±30° of the camera’s line of sight.
- Clear Field of View: No permanent shelves, signs, or screens blocking the lens. If the environment changes regularly (e.g., items restocked or large displays added), re-check the camera’s view periodically.
- Appropriate Distance: Ideally, visitors stand within ~4 meters of the camera. Beyond 4 meters, faces risk being too small (less than 60 pixels in height).
In a good installation, you’ll see:
- People’s faces are framed frontally in the camera view.
- Lighting is even, with minimal shadows.
- The camera remains stationary (no autofocus or shaky mounts) for crisp images.
2.2 Incorrect Placement
-
Camera Too High or Too Low
- Example: A camera mounted near the ceiling at 4+ meters high, angled straight down. Faces become heavily distorted, making detection/recognition difficult.
- Another example: A camera mounted around waist level, capturing mostly torso or partial face.
-
Blocked View
- Example: Placing the camera behind a shelf or sign, assuming the lens is unobstructed—only to find real customers are hidden by the display.
-
Extreme Angles
- Example: A camera on the far left or right of a traffic flow, capturing mostly side profiles. With High Sensitivity off, faces never appear frontal enough to be recognized.
-
Far Distances
- Example: Installing a standard 720p camera in a large open area where visitors pass 6–8 meters away. Their faces become too small and fail the minimum size threshold.
3. Common Pitfalls & How to Avoid Them
Below are the top pitfalls that lead to suboptimal results, along with concrete strategies to prevent them.
3.1 Obstructions Over Time
Pitfall: You install the camera on day one when shelves are empty or arranged differently, so everything seems fine. But later, additional products, signs, or promotional stands end up blocking the camera’s field of view.
Recommendation:
- Periodically verify the camera feed, especially if store layouts change frequently.
- Train store staff or local technicians to do a quick check after major re-stocking or display changes.
3.2 Poor Lighting or Harsh Backlight
Pitfall: A bright window or spotlight behind the person causes the face to appear in silhouette (too dark), or flickering neon signs cause drastic lighting shifts.
This is an example of bad lightning conditions (face is not recognizable)
Recommendation:
- Try to ensure consistent, stable lighting in front of the face (not behind it).
- If you must deal with a bright background, consider repositioning the camera or adding soft fill-light in the foreground.
- Avoid reflective surfaces (mirrors, polished metal) in the immediate camera view—these can cause glare or false detections.
3.3 Incorrect Distance or Face Size
Pitfall: Installing the camera such that people are often more than 4 meters away, causing faces to be too small (under the recommended 60-pixel height).
Recommendation:
- For typical store aisles or kiosk zones, place the camera within 2–4 meters of where people stand.
- In larger spaces (e.g., big atriums), consider using a long-focus lens or a 4K camera—but note that higher resolution often requires more processing power.
3.4 Overlapping Camera Views
Pitfall: Two or more cameras see the same new visitor simultaneously, causing potential double counting until the system synchronizes data. Or each camera might capture partial or profile views, complicating re-identification.
Recommendation:
- If multiple cameras are necessary, try to stagger their coverage or have them focus on slightly different zones so visitors enter one camera’s range at a time.
- P2P Synchronization typically resolves duplicates, but there’s a small overlap window. Be mindful of physically placing cameras to minimize truly concurrent views.
3.5 Very Short Contact Times
Pitfall: If visitors flash by in under 2 seconds, the system may never fully register or track them. This leads to incomplete or lower-quality recognition.
Recommendation:
- Place cameras near areas where visitors naturally pause—like interactive displays, counters, or entrances with slower foot traffic.
- If short contact is unavoidable, enabling High Sensitivity might capture more partial glimpses, though it can also raise false positives.
3.6 Faces in Posters, Displays, or Mirrors
Pitfall: The system identifies static faces in advertisements or on other screens, mixing them into real visitor counts. Also, mirrors can lead to detecting the reflection of a face, adding confusion.
Recommendation:
- Avoid pointing cameras directly at posters or digital displays showing human faces.
- Use software-based exclusion zones, if available, to ignore certain regions of the image (e.g., a poster hung on a wall in view).
- Remove or reposition large reflective surfaces that might bounce a face image back to the camera.
3.7 Incorrect Software Settings or Licensing
Pitfall: FR (Face Recognition) is not enabled in your configuration, or the license expired, resulting in no data being captured. Also, some setups block the network ports needed for device synchronization.
Recommendation:
- Always confirm Face Recognition is enabled in the player’s software and that your license is active.
- Check your firewall or network policies to ensure the system can communicate with the analytics server or other cameras as needed.
- Regularly update to the latest software versions to stay compatible with any new features or fixes.
4. Putting It All Together: Good vs. Bad Summary Table
Aspect | Good Practice | Bad Practice |
---|---|---|
Height & Angle | 1.5–2.5 m, frontal view, ±30° from center | Mounted near ceiling/floor, steep tilt down/up |
Distance | ≤4 m for standard HD cameras | 6+ m away without higher-res lens |
Lighting | Even ambient light, minimal glare/backlight | Strong backlighting, harsh shadows, flickering neon |
Field of View | Clear path, no new obstructions (verify after store changes) | Blocked by shelves, signage, or later-added displays |
Multiple Cameras | Slightly staggered coverage zones, P2P sync enabled | Overlapping fields with simultaneous first-time detection |
Short Contact Times | Placement near areas where visitors pause 2–5 seconds | Busy walkway with quick passers-by, no dwell time |
Faces in Posters | Use software exclusion or avoid placing sensors toward ads | Posters with large faces directly in camera view |
Software Setup | FR enabled, valid license, correct network config | Disabled FR, expired license, blocked network ports |
5. Recommendations for Avoiding Pitfalls
-
Plan Thoroughly
- Before drilling mounts, test the camera feed in actual store lighting.
- Check coverage for typical visitor positions: do people stand close enough, long enough?
-
Regular Maintenance
- Perform periodic “camera feed” checks after merchandise rotations or promotional updates.
- Confirm lighting remains adequate throughout the day (morning vs. afternoon vs. evening).
-
Coordinate Multiple Sensors
- Use slight coverage overlaps rather than total duplication.
- Ensure your network supports P2P or central synchronization to unify visitor counts.
-
Choose the Right Mode
- If your store sees a lot of quick passersby, High Sensitivity can capture partial faces but note possible trade-offs in precision.
- For slower foot traffic or a kiosk environment, standard sensitivity helps maintain more accurate matches.
-
Leverage Software Tools
- Configure exclusion zones if a static face poster or reflection is unavoidable.
- Make sure your FR license is active and up-to-date.
6. Conclusion & Next Steps
Installing cameras for audience detection is about balance: achieving clear, unobstructed, and well-lit views while staying mindful of store layouts and traffic flow. By addressing these common pitfalls—from height and angle to lighting and overlapping sensors—you’ll significantly improve detection accuracy and the value of your audience analytics.
Ready for the Practical Guide?
In the next (and final) article, we’ll give you a step-by-step installation checklist that walks you through hardware setup, software configuration, and final testing to confirm your cameras are capturing top-quality data.
Now you can go to Article 3: Installation & Setup Checklist
Comments
0 comments
Please sign in to leave a comment.