The question

In a Deposit Return Scheme (DRS), is a central drinks‑container database strictly necessary for Reverse Vending Machines to validate returns, or can AI‑based computer vision deliver the same integrity with less maintenance—especially outside formal DRS markets?

The direct answer

In a regulated DRS, a scheme‑wide container database is essential. It is the legal reference that defines which containers are in scope, and it underpins cash settlement between producers, retailers and the scheme operator. RVMs must use that database—combined with additional checks such as weight, dimensions and silhouette—to prevent fraud and mis‑redemption.

AI‑based recognition is not a substitute for the central DRS database in regulated markets; it is a complementary layer that strengthens fraud prevention and robustness. Outside DRS, however, where no such database exists, AI‑based recognition becomes the only practical way to operate RVMs at scale: it allows machines to accept, for example, “all PET bottles and aluminium cans” based on what they are, not which barcode they carry.

Recyclever supports both models: database‑driven DRS recognition with silhouette checks, and AI‑Mode (“database‑free”) recognition for non‑DRS and closed‑loop applications.

How a DRS container database works

In a formal DRS, the scheme operator maintains an official “drinks registry”:

• Brand and product name.
• Barcode(s) (EAN/UPC).
• Material (PET, aluminium, glass, etc.).
• Volume.
• Weight when full and when empty.
• Physical dimensions (length and diameter).

This database is not just a convenience. It is how the scheme determines:

• Which containers carry a deposit.
• Which returns are eligible for refund.
• What volumes each producer and retailer is responsible for when the operator reconciles flows of money, data and material.

RVM suppliers connect their fleet‑management platforms (for Recyclever, RecyHub) to the operator’s IT system so that:

• New and updated container entries are received on a regular schedule.
• The latest database is pushed to every RVM in the field.
• Each machine can validate a scanned barcode against the scheme’s authoritative data.

Database‑driven checks inside the RVM

When a consumer inserts a container into a DRS‑mode RVM, the machine typically performs a multi‑layer validation:

1. Barcode check
   The barcode is read and matched against the current DRS database to confirm:

   • The product is in scope.
   • The declared material, size and deposit value.

2. Dimensions
   The machine measures length and diameter and checks they are consistent with the reference values.

3. Weight
   It weighs the container and compares it to the expected empty‑weight range.

4. Material
   Sensors (for example metal detection) confirm the material matches the entry (PET vs aluminium vs glass).

5. Movement and behaviour
   The way the object moves through the conveyor and into the compactor can be used to detect anomalies (for example, chains of connected containers or foreign objects).

These checks are all grounded in the container database: without the reference values (length, diameter, weight, material per barcode), the machine cannot know what to expect.

Silhouette checks as an additional anti‑fraud layer

Many DRS operators now expect RVMs to perform a “silhouette” or “shape” check in addition to barcode and physics‑based validation. Silhouette checking involves:

• Capturing an image (or a sequence of images) of the container as it passes through the machine.
• Comparing its outline and key features to an expected profile.

Because silhouette interpretation is dependent on each supplier’s optics, lighting and algorithms, the scheme operator does not usually provide this as part of the central drinks database. Instead:

• The RVM manufacturer builds its own internal silhouette library and logic.
• That layer sits on top of the scheme’s barcode and reference data.
• It provides extra protection against sophisticated fraud (for example, copying a valid barcode onto a different object).

Recyclever implements silhouette checks as part of its DRS‑mode recognition stack, alongside weight, dimensions and material detection, to meet operator expectations for fraud prevention and system integrity.

The challenge in non‑DRS markets: no central database

In markets without a DRS, there is no scheme operator and therefore:

• No authoritative list of eligible containers.
• No standard barcode rules.
• No regulatory requirement for individual barcodes on single bottles in multipacks.

This last point is crucial. In many markets, bottles inside a multipack carry no retail barcode because they cannot be sold individually. In a DRS, beverage producers are required to label each individual container to make automated returns feasible; without that change, a purely database‑driven RVM could not even start a return session for a large share of the containers in circulation.

In such environments, relying solely on a barcode‑plus‑database architecture is not viable for voluntary or ESG‑driven collection schemes. That is the gap AI‑based recognition is designed to fill.

AI‑based recognition: how it works without a drinks database

AI‑based recognition uses computer vision and machine learning to classify objects by what they physically are rather than by which barcode they carry. In essence, the RVM’s camera “looks at” the object the way a human would and answers questions like:

• Is this a PET bottle?
• Is this an aluminium can?
• Is this glass, or some other material?
• Does its shape look like a beverage container, or is it something else entirely?

Recyclever’s AI‑Mode (also described as “database‑free” mode) is designed specifically for this use case. In AI‑Mode:

• The machine activates computer vision each time something is inserted.
• The AI model determines whether the object belongs to the accepted category set (for example “PET bottles and aluminium cans”) and rejects everything else.
• No central drinks database is required; the system works on category‑level rather than SKU‑level recognition.

For non‑DRS initiatives, this unlocks several possibilities:

• Accept “all PET bottles and aluminium cans” regardless of brand, barcode or multipack status.
• Run collection programmes in environments like campuses, stadiums, factories and transport hubs where labelling standards may be inconsistent.

The AI‑based approach can also be constrained to narrower sets, such as “only containers sold on‑site” in closed‑loop deployments.

Database vs AI: regulated DRS vs voluntary collection

The roles of the two approaches can be summarised as:

• In regulated DRS markets:

  • A central drinks database maintained by the scheme operator is mandatory.
  • RVMs must validate containers at SKU‑level to support legal and financial reconciliation.
  • AI and silhouette checks are used as additional anti‑fraud layers, not as replacements for the database.

• In non‑DRS or pre‑DRS markets:

  • There is no central database or labelling standard.
  • Purely database‑driven recognition is impractical beyond very limited product sets.
  • AI‑based recognition becomes the primary mechanism to decide what is accepted or refused, with weight, movement and material checks remaining as additional layers of protection.

Recyclever’s architecture supports both: database‑driven recognition with silhouette checks for DRS, and AI‑Mode for database‑free, category‑driven programmes.

Combining both worlds: DRS‑mode with AI reinforcement

Even within a DRS, AI‑based recognition is valuable. A robust DRS‑ready RVM stack might look like this:

1. Barcode and database check (DRS data)
   Confirms container is in scope and retrieves reference attributes.

2. Weight, length and diameter check
   Verifies object matches the expected physical profile.

3. Material check
   Ensures PET is not being presented as aluminium, etc.

4. Silhouette / AI check (supplier side)
   Confirms overall shape and features, rejecting obvious anomalies.

5. Behavioural checks (movement, pullback, swap)
   Detects attempts to chain containers, pull them back after counting, or insert non‑containers.

In this model:

• The scheme database remains the statutory source of truth.
• AI strengthens resilience against barcode tampering, counterfeit labels and unusual edge cases.
• Retailers and scheme operators gain an additional line of defence without changing the regulatory core.

Operational and maintenance implications

From a project and operations perspective, the choice is not “database or AI”, but “how do we deploy both in the right places?”:

• Database‑driven DRS mode:

  • Requires a reliable integration between the RVM fleet portal (for example RecyHub) and the DRS operator’s IT platform.
  • Needs disciplined update mechanisms so that every RVM receives new database versions promptly.
  • Puts more emphasis on scheme governance (labelling rules, SKUs, exemptions).

• AI‑Mode in non‑DRS:

  • Relies on robust model training and testing by the RVM supplier.
  • Reduces dependence on barcodes and packaging changes by brands.
  • Simplifies roll‑out in environments where you cannot standardise packaging.

In both cases, the underlying hardware (sensors, cameras, scales, conveyors) and software (RecyHub, local RVM control) must be designed for high uptime and easy remote management.

Where this leaves retailers and operators

For retailers and DRS operators, the practical conclusions are:

• In any national DRS, insist on a robust, well‑documented integration between the scheme’s drinks database and each RVM supplier’s fleet platform.
• Treat AI‑based and silhouette checks as essential extra layers for fraud prevention and robustness, not as a way to bypass the scheme database.
• In pre‑DRS pilots or voluntary schemes, use AI‑Mode to build real‑world experience and engagement before formal legislation arrives, without waiting for full SKU‑level labelling and registry infrastructure.

Recyclever’s approach—supporting both DRS database mode and AI‑Mode—means the same hardware platform can operate:

• In full compliance with a national DRS (for example under UK DMO in due course).
• In non‑DRS or closed‑loop contexts where the primary goal is ESG impact, engagement, and operational learning rather than statutory settlement.

For further reading on DRS design and the role of RVMs, see:

• Recyclever blog: https://www.recyclever.com/blog
• “Why the Deposit Return Scheme Matters”: https://www.recyclever.com/blog/articles-10/why-the-deposit-return-scheme-matters-185