New Type Of Coin Sorting Machine

Building a robot coin dealer? Impossibly hard. Identifying the date and mintmark is child's play for a human but tricky for a robot; you'd need to have some kind of OCR scanner to first find the date, then read it. Then find the mintmark and read that.

Grading would be the toughest. Computerized coin grading has been attempted ever since the 1980's; it's never worked well. The "aesthetics" and "eye appeal" component of grading is too high; they're the sort of things that computers are notoriously bad at detecting.

Sorting would also be tricky; you'd need thousands of catchtrays to catch each and every grade of each and every date/mintmark of each and every denomination.

To calculate numismatic value, it would have to have access to a database of coin values; this would be the easiest part.

What would the robot do with the numismatic coins you didn't want to keep?

you know whats funny. That is the exact idea that got me into coin roll hunting last year. I called a coin shop and asked the guy if silver still circulated. He said yes but only look in halves. So I thought yea right I'll go do a lil research at a few banks before I start designing my machine. So thanks to that coin guy I scored like 150 40% and like 11 90% on my first day. I even got my teller to help me sort the silver out at the counter. The teller next to us asked "are you looking for silver you wont find any". so I politely hold up a mixed stack of silver and say you mean like these lol. The look on her face was priceless :) anyway since that day I've been hitting banks, going through boxes, and working on my sorting machine. Right now it uses (DR.Evil voice)-> "electromagnetism" to id the coins I want to sort out. Cents through halves. Some of you will know what device is the brains of most sorters. But if you have tried to build your own you know its the hopper and delivery system that's a you know what to incorporate. on a side note when I have my machine set for silver halves it will go through a box in 6 min :)

So back to "our" idea for visually sorting out the coins via a machine. All you need to do is feed one coin at a time under a digital camera, capture the image, flip over the coin and capture the other side. those pics go into photoshop where you can setup a batch process to rotate the images vertical. Then use its or another software to id the date and mint mark. As far as grading I agree with the above posts that it would not work that well. but I think it would have some luck with certain types of errors :)

No, you are looking at it too simplistically. Eye appeal is purely subjective to the viewer, software could never account for that. The sharpness of the strike, the flash of the luster, and the patina all go into eye appeal. Sure, luster is a type of reflectivity but a polished coin is pretty darn reflective too. As for grading, it is not technologically feasible. Various companies have poured huge resources into it(including PCGS which currently holds a few patents) and it has been an abject failure.

But software could look at the wear on the coin in conjunction with the reflectivity and determine a coin to be "cleaned" if it is too reflective for the wear patterns. Software could also look for patterns that determine if a coin has been cleaned.

Well I'm sure the first machine would be extremely expensive. Developing the hardware and software would take time and money. You would also need to get comparisons for the software to learn how to grade. Then the testing, the UL listing.. there is a lot to go into it.

But after you have a successful prototype and you can break down the process for reproduction... It's cost would be a lot less. The machine would have to have an embedded modem to receive software updates and also to let you know when the bins are getting full (just like coinstar does now)

Sifting silver out of a given denomination is "easy" with a programmable coin comparator. It's the same principle people have been using to separate copper cents from zincolns. I put that in quotes because unless you are are very mechanically inclined, building the feeding mechanism is a difficult task. So sorting the denominations and even the metals is a task that has been solved in various ways (vending machines and coin stars do this all day).

If you were going to set out to develop such a machine, your best bet is to start with the software aspect. You don't need any specialized equipment, just a scanner to start with. In theory a high speed camera would take an image similar to a scan. You can assume that before scanning a given coin you already know the denomination because it's been mechanically sorted and just go from there. There's probably a photoshop filter or other image algorithm that can orient the scan so that the writing is in the proper orientation. Scanning the year and motto should be doable with standard OCR packages. From that you can determine the rough area where the MM should be and use OCR to read that (there will be cases were that doesn't work).

The hard part is determining the grade - software might be able to get you a rough estimation with facial recognition type algorithms, but you won't be able to differentiate between MS62 and MS63 for example or even between AU and MS. Might not even be able to tell much beyond a G or an AU. You should be able to differentiate between coins of the same year with different designs (1909 cent for example). This would require some serious knowledge of image processing though. Presumably if you got that all figured out you could pay someone to design the mechanical aspects of the machine.

So, I think you'd need to be a mechanical engineering expert and an image processing genius to build such a machine and at best you'd be able to sort the silver out, spit out any coins that have interesting date/MM combos and dump all the rejects somewhere for human review. It would also be extremely slow and expensive.

You're probably better off spending all that time roll hunting :)