AI product recognition in self-checkout: supplement or replacement for the barcode?

First at traditional checkout lanes, then at self-checkout stations: barcodes have shaped retail checkout processes for decades. But with computer vision and AI-powered product recognition, new possibilities are emerging for self-checkout workflows and frictionless shopping experiences. These systems are designed to reduce scanning errors, automate processes, and make checkout more convenient.

To this day, the barcode remains one of the most efficient methods of product identification in retail. Products can be captured quickly, in a standardized way, and with a high degree of accuracy. At the same time, barcodes are inexpensive to implement, require very little computing power, and integrate easily into existing merchandise management and POS systems.

These advantages play a particularly important role in brick-and-mortar retail, especially in checkout processes. The POS system does not actually “read” the product itself, but only its unique identifier, which is then used to retrieve the stored product data. This allows the process to function independently of a product’s visual appearance and enables fast, standardized identification across a wide range of retail operations.

Many self-checkout processes using self-scanning kiosks or smart carts also continue to rely on barcodes. Customers scan items themselves at the self-checkout station, directly on the shopping cart, or with mobile devices. This reduces wait times and creates a more flexible, self-directed shopping experience.

At the same time, however, this also highlights the limitations of traditional barcode-based processes: every item must be actively scanned. Missed scans, unreadable barcodes, or forgotten items can interrupt the process or require additional control mechanisms.

Visual product recognition is also particularly interesting for items without a traditional barcode – such as fresh produce like loose fruits and vegetables. When these products are recognized by AI, they can be suggested directly to customers on the self-scanning screen.

In visual product recognition, cameras analyze product images using computer vision and deep-learning technologies. The system must clearly identify specific items and product variants – in retail, this is referred to as “SKU-level recognition.”

Technically, this is based on neural networks that analyze visual features such as shape, colors, textures, lettering, or packaging design and convert this information into mathematical feature vectors.

AI-powered product recognition in brick-and-mortar retail is considered significantly more complex than traditional object recognition because the systems must handle several challenges simultaneously. Retail assortments often contain thousands of products that need to be distinguished from one another. On top of that, there are highly similar product variants (for example, different sizes or flavors), constantly changing packaging designs, and seasonal or newly introduced items. As a result, AI models must be continuously retrained and updated.

In addition, cameras do not always have an ideal view of products. Items may be packed closely together, partially obscured by other products or customers’ hands, or captured from different angles.

AI-powered product recognition can support loss prevention in self-checkout environments. The system analyzes the checkout area through cameras and detects when items are moved but not properly scanned. The missing item can immediately be displayed on the customer screen and suggested for correction. In this case, product recognition serves as an additional verification layer within the checkout process.

AI-powered product recognition can also replace the scanning process entirely, for example in a smart cart. Cameras built into the shopping cart automatically recognize products as they are placed inside and add them to the shopping list shown on the screen or in the app. Removed items are automatically deleted from the digital cart again.

The differences between traditional barcode scanning and visual product recognition become especially apparent during the checkout process: