Counting pallets in a warehouse using video analysis involves applying a computer vision model to the feeds from cameras that are already installed, in order to detect pallets in defined areas—loading docks, buffer zones, storage aisles—and generate a continuous count. Unlike scan-at-receipt or RFID, this method does not track an individual pallet: it measures the quantity present at a given location and time.

This is precisely where it fills a gap. Most warehouses know what has come in and what has gone out, because these events are scanned. Far fewer know, at 3:00 p.m. on a Tuesday, how many pallets are actually in the buffer zone or how many dock positions are occupied. That level of visibility isn’t found in the WMS—it’s observed by walking through the warehouse.

Why does pallet counting remain a blind spot?

The warehouse information system operates in terms of events: a pallet is received, assigned to a location, picked, and shipped. Between these events, its actual physical location is based on convention rather than observation.

Yet discrepancies accumulate precisely in the areas that aren’t scanned: the clearance zone in front of a loading dock, the buffer area before quality control, the aisle where items are set aside “for now.” These areas absorb the fluctuations of daily operations, and this is where congestion builds up.

The usual solution is a rolling inventory or a zone inspection conducted by a team leader. While reliable, this method is only a snapshot: it provides a figure at 8 a.m. that says nothing about the situation at 2 p.m., when a decision on dock assignment would have been useful.

What methods can be used to count pallets?

Scanning and the WMS

This is the gold standard for item-level traceability: each pallet is identified, linked to an order, and tracked throughout its cycle. The downside is that the data only exists at the moment of scanning, and any unscanned handling becomes invisible.

RFID

RFID tags enable contactless reading and much faster inventory taking. However, this model assumes that pallets are tagged—which raises the issue of incoming pallets not tagged by suppliers—and that gate readers or mobile readers cover the relevant areas.

Position Sensors

Floor sensors, photoelectric sensors, and location systems mounted on forklifts: while effective within a limited area, they require hardware deployment proportional to the covered surface area, which limits their scalability to an entire site.

Video analysis

Warehouses are already widely equipped with cameras, installed for security and to resolve disputes. Video analytics repurposes this footage to generate a count by zone, without requiring additional hardware or modifying existing scanning processes. It does not replace the WMS: it documents what happens between scans.

How exactly does video analysis count pallets?

The process consists of three steps. An object detection model identifies pallets in the image, including those that are partially obscured or stacked. Areas of interest are then drawn on each camera’s view to define what is being counted: a specific dock position, aisle, or floor section. Finally, the count is reported at regular intervals, with a history that allows you to reconstruct the area’s occupancy curve.

Two metrics are derived directly from this data. The occupancy rate of an area is calculated by dividing the number of pallets present by the area’s theoretical capacity. The dwell time measures how long the area has been occupied beyond a certain threshold, which identifies bottlenecks more effectively than a single snapshot.

What technical conditions are required for reliable counting?

Reliability depends less on the model than on the scene being observed.

The viewing angle. A low-angle camera obscures the pallets located behind the front ones. A high-angle view, even a partial one, distinguishes the units much better. In an existing facility, repositioning one or two cameras per zone is often enough to improve the quality of the results.

Lighting and obstructions. Dock areas alternate between backlighting and dim lighting depending on the time of day and when doors are open. A forklift parked in front of a stack of pallets can permanently block the count. These issues are addressed during calibration by combining multiple views or excluding certain time periods.

Variety of pallets. Whether wrapped in film, partially loaded, on Euro pallets, or in custom sizes: the greater the variety, the more critical the adjustment phase becomes. This is the primary factor causing discrepancies between a demonstration and a real-world site.

Defining what is being counted. Which zone does a pallet straddling two zones belong to? The rule must be established with the operations teams before the system goes live; otherwise, every discrepancy found will reignite the discussion.

What is the purpose of this daily count?

Three uses come up most often.

Managing dock congestion. Knowing in real time which positions are available allows you to assign a truck upon arrival rather than on a first-come, first-served basis, and to anticipate a bottleneck before it spreads to the yard.

Objectively assess buffer zones. An occupancy trend over several weeks reveals whether a zone becomes saturated occasionally or structurally—two issues that require different solutions, and which are often confused in practice.

Detect discrepancies with the system. A physical count that differs from the theoretical inventory in a zone signals an unscanned transaction or an allocation error, well before the next inventory.

What is the legal framework for filming work areas?

A warehouse is a workplace, which changes the situation compared to a space open to the public. The CNIL strictly regulates video surveillance in the workplace: cameras must not place employees under constant surveillance, and a system cannot be used to monitor an individual operator’s activity.

Counting pallets remains within these guidelines under three conditions: it focuses on objects rather than people; it does not generate any personally identifiable information or individual performance data; and its scope is limited to relevant logistics areas. In addition, there are obligations specific to the workplace—informing employees, consulting with employee representative bodies, and registering the processing activity in the data processing registry. These matters fall under the responsibility of the operator’s DPO and human resources department and should be addressed before deployment rather than after.

Where to start?

The quickest route is to start with a single area and a single metric. A loading dock area or buffer zone covered by two or three existing cameras, a continuously updated count, and two to three weeks of comparison with the team’s manual counts. This calibration phase establishes the acceptable margin of error and, above all, builds confidence in the data—a prerequisite without which no metric will be used to make decisions.

XXII is developing CORE, a French video analytics platform that leverages existing cameras to generate these types of warehouse metrics: zone-based counts, occupancy rates, dwell times, and threshold alerts. This approach avoids the need for additional hardware and integrates with existing operational tools, within a framework that complies with the GDPR.