Massive data matrix image processing in Pharmaceutical Logistics
Summary of the entity:
Cofares is the leading pharmaceutical wholesaler in Spain. It is a cooperative with 100 % pharmaceutical capital that provides pharmacies with products as well as health related services.
Summary of the challenge:
Due to the implementation of the FMD (Falsified Medicines Directive) all the medicines packages must be identified with a unique serial number embedded in a data matrix code printed on the package.
There are some situations that require the wholesaler to verify the status of the serial numbers in an European repository.
Currently this verification is done by scanning a data matrix in each package one by one. All packages are stacked into a carton case and an operator opens the carton case and takes each package by hand to scan its data matrix, spending a long time in the process.
Cameras are being installed to take images of the open cases at a workplace so that a software program processes them and extracts all the information from the data matrix.
We have the following limitations:
- We can’t use very powerful computers (e.g. expensive GPUs).
- There is no internet connection (images can’t be sent out the computer).
- The photos are taken with a 12Mpx (lower resolutions are enough to read the data matrix).
- We expect very low AI compute times (less than 1 second per image).
We also want the software to detect if the image is not capturing the full layer (for example, the flap of the carton box covers medicine boxes or there are some reflections) to warn the operator about it.
To automate the process of scanning so that the operator stops doing it for most of the packages and only supervises and solves specific problems.
Any improvement suggestions about the physical environment and the devices used (camera, computer, lightning, etc.) will be welcome.