Object Recognition for Inventory. Where to start?


I'm work in the tool room of a CNC machining company. One of the biggest headaches in the job is identifying misplaced "inserts" for tooling. These inserts are about a square inch at the largest. We have about 200 inserts to manage. The problem is that, when we replace an insert it usually ends up in a pile of "good inserts" that need a home. However, identifying each little insert is not something anyone has committed to memory so they often just sit in a pile.

I'm looking to build a device that can identify an insert via camera object recognition, then give the insert name and provide the location they go. These inserts are small and have various shapes (square, triangular), designs on the face of the insert, and specific radiuses on the edges. The Pixy looks neat, however it seems to be heavily color based. These inserts are typically browns, blacks, greys, and golds and some look almost identical except fine edges. I would need a highly detailed camera that identifies small shapes.

Is this do-able with the arduino Uno or would I have to go full Raspberry for this project? Are there any available resources for a similar project? I'm not looking to "track" or "follow" anything, simply identify the object and output relevant information for sorting. Where do I start?

IMO go for a Pi or stronger.

For static image recognition, I've used a number of "Blob detection" libraries with great success.

It'll be tedious building a functioning database, but I imagine you could get it working.

Sounds like you could also CNC write a ID number on each.. if you only had a CNC machine around in that CNC shop...

The reality of “the pile” in a workshop just means the most used items are on top of the pile and the least used are at the bottom.

Not a job for an Arduino. Image recognition is outside of its realm.

What probably works best is teach your colleagues to return the inserts immediately to the correct location. Saves a lot of work in the long run (in terms of sorting and searching).