A matching total can hide counting errors
Hypothetical example: 20 boxes pass and the screen shows 20. The system may still have missed two boxes and counted two others twice. Errors cancel out in the total. Review individual events before deciding whether the count is useful.
This is a proposed check for a warehouse or small manufacturer, not an Adaris measurement or accuracy guarantee. Start with a specific decision: checking an outbound movement, reviewing a tray or flagging a difference for a person to investigate.
1. Define what counts as one box
A visible box is not necessarily a box that has left the warehouse. Agree a rule, such as crossing a line in one direction. Decide what happens when a box stops, moves back or crosses again. Visible units, crossings and completed orders are different measures.
- Object: individual box, tray or pallet.
- Event: presence in a zone or crossing a boundary.
- Exceptions: reverse movement, stops, returns and several units moving together.
2. Include difficult conditions in the sample
Select representative clips and review them manually. Include quiet and busy periods, shadows, lighting changes and partial obstruction when they occur in the actual process. Do not select only clear footage.
Reserve different material for the final check. Repeatedly tuning against one clip only demonstrates improvement on that clip. Agree how footage will be shared and who may access it first; the initial enquiry can describe the scene without attaching video.
3. Record three kinds of error separately
For each clip, keep its time or reference, the manual count, the system count and the reviewed events. Record missing boxes, repeated counts and detections of objects that were not boxes. If the footage is inconclusive, mark the event as uncertain.
In the 20-box example, report two omissions and two additional counts even though the difference between totals is zero. If checking a shipment is the objective, also verify which events belong to that shipment: a counter alone does not identify an order.
4. Agree what is useful for your operation
Set acceptance criteria before testing: which errors matter, the conditions to evaluate and who reviews exceptions. Measure correction time as well. A convincing number is of little use if someone still has to watch all the footage.
The outcome may be to proceed, change the camera view or stop. A poorly positioned camera may need a physical adjustment before more software. Adaris can quote a feasibility test with an agreed sample, deliverables and limits. Production deployment and warehouse-system integration are scoped separately.
Related services
What do you count manually today?
Describe the object, where it moves and why you need the count. Adaris works from Elche with SMEs in Alicante and across Spain; commercial discussions are currently in Spanish.