Stock counting is one of the most labour-intensive, error-prone, and disruptive activities in any UAE warehouse or store. Computer vision is changing this — cameras and AI that count inventory automatically, continuously, and feed the results straight into Odoo. This is where Rubicon’s two specialisms — computer vision and Odoo ERP — combine most powerfully.
The Problem with Manual Stock Counting
- Full physical counts require shutting down operations or working overnight
- Manual counts are error-prone — miscounts, transposition errors, double-counting
- Counts are point-in-time; the moment they’re done, they start drifting from reality
- Reconciling count variances against Odoo is time-consuming
- Cycle counting helps but still consumes significant labour
How Computer Vision Stock Counting Works
Several complementary approaches, depending on environment:
Fixed Camera Counting
Cameras mounted over storage areas continuously monitor stock levels. AI counts units on shelves or in bins and compares against the Odoo expected quantity. Variances flagged in real time.
Mobile / Handheld Vision Counting
A staff member walks the aisle with a camera device (tablet or dedicated scanner). AI counts items in view as they pass, dramatically faster than manual counting and feeding directly to Odoo.
Drone-Based Counting (Large Warehouses)
For high-rack warehouses, drones with cameras scan upper levels that are otherwise hard to count, reading location labels and counting pallets.
Receiving and Dispatch Counting
Cameras at dock doors count items in and out automatically, keeping Odoo inventory accurate at the points of highest movement (see our warehouse AI article).
Integration with Odoo Inventory
The AI counting layer connects to Odoo Inventory:
- Counts written back as inventory adjustments (after human confirmation)
- Variances flagged against Odoo expected quantities
- Continuous counting replaces or augments scheduled cycle counts
- Stock accuracy dashboards in Odoo show real-time confidence per location
Where This Works Best Today
| Environment | Suitability | Notes |
|---|---|---|
| Retail shelves (uniform products) | High | Shelf-out detection mature |
| Warehouse dock (receiving/dispatch) | High | Carton counting reliable |
| Bulk/palletised storage | Medium-High | Pallet counting works well |
| High-variety small parts bins | Medium | Improving; depends on item distinctiveness |
| Loose/irregular items | Lower | Hardest case; depends on context |
The Accuracy Reality
Computer vision counting accuracy depends heavily on the environment. For well-suited cases (uniform cartons, distinct products, good camera angles), accuracy is high enough to reduce manual counting dramatically. For difficult cases (mixed small parts in deep bins), it augments rather than replaces manual processes. A realistic deployment combines vision counting for the 70-80% of inventory it handles well with targeted manual counting for the rest.
The Business Case
- Reduce or eliminate full physical counts (and their operational disruption)
- Cut cycle-counting labour by 50-80% in suitable environments
- Keep Odoo inventory accurate continuously, not just at count points
- Catch shrinkage and errors days or weeks earlier
- Free staff from counting for higher-value work
Edge Deployment for UAE
Consistent with Rubicon’s edge-first approach, vision counting runs on local edge AI hardware (NVIDIA-supported edge AI) — video stays on-premise, data sovereignty preserved, no cloud dependency, low latency.
Implementation Approach
- Identify the inventory zones where vision counting suits best (start with receiving/dispatch or uniform retail shelves)
- Deploy a pilot in one zone — cameras, edge server, Odoo integration
- Run vision counting in parallel with manual counting to validate accuracy
- Once validated, reduce manual counting in that zone
- Expand to additional zones based on suitability
Free 30-minute warehouse vision assessment.