Pipe counting bot with photo-based object detection

Сomputer vision
Issue
Inventory management is not simple, especially if it is done manually in a warehouse. It becomes even more challenging during the transportation, which requires revising of waybills. In particular, this becomes harder when the goods are transported without packaging, for example, pipes which can be easily lost or stolen.
We developed a Telegram bot that makes photo-based calculations of pipes' count in a truck or at a warehouse.
Solution
Objective
Automate the routine: simplify the calculation of pipes when transporting from and to warehouses.
For whom?
01/
For everyone who works with pipes in the fields of
Transportation
Production
Construction
Other purposes
02/
Any other items can be
counted by this method
logs, tie stays, piles, beams, etc.
Additional model training is required
No need to use the Telegram-bot only, a web interface can be used, as well
How does it work?
03/
A warehouse specialist takes a picture of pipes with sections being seen, as the model uses them for object calculations.
They send the photo to the Telegram bot
The model recognizes the number of pipes and
sends the number back to the specialist.
If needed, the specialist can edit the number, but usually it is not a case.
What we have done
04/
Integrated it into the bot to enable the client-product interaction
Made mark-ups and trained the model using real data
Collected client data as all pipes have different diameter, material, etc.
Difficulties
we faced
05/
Issues
Large amount of pipes which are all of different kinds
Cameras with different qualities, which can make some photos blurry
Visual shading and overlapping in the photo. Sometimes pipes are placed chaotically overlapping and shading each other.
A huge load of data was needed, as well as very accurate mark-ups
Single solution

Tools

06/
zDialog фреймворк

Detectron2

Library for object recognition
OneDash сервис
Details

Markup

Internal mark-up tool
Contact us