Resolution
At first glance, it may seem that any camera will do, but in practice that’s not the case. If the camera resolution is below Full HD (1920×1080), the system begins to confuse small details, cannot see price tags clearly, and struggles to distinguish between similar products. If the products are small or the shelf is long, it’s better to use cameras with higher resolution.
Saving money on this parameter often results in a loss of functionality for the entire system.

Field of View
A field of view that is too wide creates a “fisheye” effect: objects near the edges become stretched and products appear distorted.
A field of view that is too narrow means part of the shelf will remain outside the frame.
It’s important to choose a lens so that the entire required area is clearly visible without distortion at the edges.

Mounting Height and Placement
The camera should view the shelf at roughly the same angle as an average shopper.
If cameras are installed too high, you get a top-down view where products overlap each other. As a result, a significant part of the assortment becomes invisible to the system.
If cameras are installed too low, you may run into focus problems with the upper shelves.
For island shelves or refrigerators, custom installation setups may be required, and it’s best to test these configurations in advance.

Capture Frequency
The capture frequency depends on the situation. If you don’t need to track changes every minute, one image per hour is often enough.
However, if it’s important to detect instant out-of-stock situations or respond quickly to changes in product placement, more frequent capture may be required — even close to real time.
Keep in mind: the higher the capture frequency, the greater the load on data storage and processing.

Sometimes lighting matters more than the camera itself — and this is exactly one of those cases. If the lighting is poor, even a high-end camera becomes ineffective. A £10,000 camera in bad lighting can deliver worse results than a £200 camera in good lighting.
Lighting varies significantly across different store sections, which means the same product can look completely different depending on where it’s placed.

It Changes How Products Look in Images
Poor lighting alters how products appear in photos. Colors get distorted, contrast either drops or becomes too harsh, and small text on packaging becomes blurry.
In dim lighting, red can look brown, and blue can appear almost black. As a result, the system starts confusing similar products. In practice, there have been cases where a can of cola in a refrigerator and the same can on a regular shelf were recognized as entirely different items.

It Creates Shadows, Overexposure, and Glare
Improper lighting makes it harder to see the product as a whole.

• Shadows can cover parts of the shelf. Products on upper rows cast shadows on those below, and items deeper in the shelf can end up almost completely in the dark. The system only sees fragments of the packaging — not enough to identify the product accurately.
• Overexposure occurs when bright light hits the shelf directly. Overexposed areas turn into white patches with no detail, which is especially common near windows on sunny days.
• Glare is a major issue with glass surfaces. Refrigerators with transparent doors reflect light like mirrors. Instead of the product, the camera captures reflections. The glass may be clean and the product perfectly placed, but due to glare, the system effectively sees nothing.

All of these issues degrade image quality — and the worse the image, the worse the recognition.

It Makes the System Rely on Guesswork
When lighting is very poor, the system starts to “guess.” It may try to compensate using assumptions like: “This shelf usually contains product X, so it’s probably that.”
But this approach is unreliable. Products get moved, displays change, and shelf layouts are frequently updated. The whole point of recognition is to achieve accuracy, not to make educated guesses about what might be on the shelf.

Failed Setup 1: Ceiling Camera in a Hypermarket
Configuration:

• 4K camera mounted at a height of 4.5 meters
• 120° field of view to cover maximum area
• One camera covering three double-sided shelving units (around 15 meters of shelves)

Result:

• Upper shelves partially block the lower ones
• Products appear as colored stripes, with no visible detail
• The system recognizes only part of the ассортимент
• Price tags are completely unreadable

Reason for failure:
Trying to save on the number of cameras. Covering too large an area with a single viewpoint led to a loss of data quality.

Failed Setup 2: Refrigerators with Internal Lighting
Configuration:

• Camera mounted above the refrigerated display
• Bright LED lighting inside the fridge
• Glass doors
• Shooting angle: top-down at 45°

Result:

• Glare from glass and LED lighting completely overexposes the image
• Reflections create white patches over ~60% of the frame
• Condensation on the glass further blurs the image
• After opening the doors, fogging makes the system effectively blind for 15–20 minutes

Reason for failure:
Failure to account for the physical properties of the environment. Glass, strong lighting, condensation, and other predictable factors were ignored during system design.

Successful Setup 1: Modular System in a Premium Supermarket
Configuration:

• 5 MP cameras mounted at 1.8 meters
• Each camera covers 2–3 meters of shelving (one section)
• Field of view: 60–70°, front-facing
• Distance to shelf: 1.5 meters

Result:

• System recognizes up to 95% of products
• Even small packaging details are visible
• Price tags are clearly readable
• Accurate product counting on shelves
• The system detects even minor product shifts

Outcome:
A well-balanced approach between camera quantity and coverage quality. The investment in equipment pays off through high data accuracy.

Successful Setup 2: Refrigerators with Proper Angle
Configuration:

• Cameras mounted frontally at mid-shelf level (1.2–1.4 meters)
• Distance to glass: 0.8 meters
• Shooting angle perpendicular to the glass (0° to the normal)
• Polarizing filter added to the lens

Result:

• Glare minimized thanks to perpendicular angle and polarization
• Condensation is less critical (short distance compensates for sharpness loss)
• Recognition works in ~90% of cases
• System recovers within 3–5 minutes after doors are opened

Outcome:
Combining technical solutions for glare reduction, the correct shooting angle, and close proximity to the subject leads to strong and stable performance.