Sharp eye on the money

Written by: Tom Shelley | Published:

Tom Shelley heads for Lancashire to look at bank note and coin recognition techniques that might well add the golden touch to many automated tasks

Intelligent, multi-colour vision systems and advanced inductive sensing systems, far more sophisticated than are used in most industrial sensing and inspection systems, can now swiftly verify the validity of banknotes and coins in automated payment and gaming systems.
Strategies have been developed, in many cases using processors developed for the games industries, which allow a few kilobytes of instructions effectively to mine out information of significance from data sets that can run to terabytes.
And because these technologies have been developed for budget-conscious customers making vending and gaming machines, they have the potential to be applied widely across industry. For example, while this technology has been honed to distinguish genuine notes and coins from forgeries, it could also be invaluable in distinguishing good products from unacceptable, whether at the premises of a manufacturer or in ensuring quality control of bought-in components.
Money Controls, the company behind these and many other patented technologies, has been making money acceptance and change-giving machines for the last 40 years.
Its current banknote identifying and verifying technology uses small, quite low resolution cameras to capture digital images of notes, which unlike other machines on the market, can accept them whether they are put in at an angle, or are damaged, provided this is not in a critical area important to distinguish genuine notes from forgeries.
Based in a century-old former cotton mill in Oldham, Money Controls uses an approach that, in the words of technical director Mike Bell, is “obvious in retrospect, except nobody thought of it before”. And that is to capture images of the upper and lower faces of the currency in question, straighten and examine them, work out what value note it is likely to be, check that it is and then look for known features – features of known forgeries, in this case, or likely defects in the case of industrial products – and then either accept or reject this.
At Oldham, engineer Kevin Mulvey performed a series of demonstrations using US $20 bills, one of which had its corner folded over. In each case, the system straightened the image, identified the bills as $20 denominations, closely examined appropriate areas to confirm the values, scrutinised those areas associated with forgeries, before it accepted the notes – and all within a few seconds.
The light comes from light boxes that cycle through three RGB visible and two infrared wavelengths from rows of chip LEDs, with ingenious diffusers, which software manager Andy Barson describes as one of the harder parts to design. These diffuse light from sources quite far apart, so they come out as solid bars of light. The infrared is crucial, because many notes employ areas that absorb infra red light, instead of reflecting it, in order to easily distinguish genuine notes from photocopies. A side benefit apparently of particular interest to casino owners is that images of notes can be retained, so that if a gambler believes he or she has not been credited with the correct denomination, the image of the last note can be instantly displayed on the screen of the game machine, pulled up by a central control room or extracted on a pocket PC, rather than having to open it up to look physically. The facility can also be used to store image data relating to new forgeries and so add this to the database.
The really clever part, however, lies in the software, which has to distinguish genuine notes that may have become worn or dirty, or slightly shorter and fatter with age, from forgeries. This involves scanning in different notes in various conditions – which is where the terabytes of data come in – and working out the possible range of variations that are acceptable. We were shown images in the form of three-dimensional topographic displays, with the notes as bases and measured values as heights. “We generate an image of a mean population of good bills in 3D and frauds in 3D,” explains Bell, “and subtract one from the other, and look for the parts that make the decisive differences.”
Despite this complexity, tables of data to undertake the verification process are only a few kilobytes, and updated versions can be downloaded over the Internet for registered users, in order to help recognise new forgeries.
Coins, on the other hand, are primarily examined by inductive technology, but in a way that is much more sophisticated than is used in inductive proximity sensors, with collected data assessed using an artificial intelligence methodology to accommodate population variation.
Positions of coins in the machines are located optically, but surfaces and the constructions of the coins are examined by pairs of coils on each side of the coin – five each, in most of the company’s machines, running at a range of frequencies and phase relationships. But in what is the world’s first high-speed, multi-coin gaming acceptor with a vertical free fall, there is only one rectangular coil and a series of light beams. The coils are produced on a special, automated manufacturing line the company has developed itself, which can turn them out to values that are constant within 0.1mH and 0.05 ohms, according to manufacturing director Ken Collett. “Winding tensions on the wires are crucial,” he stresses.
The coils are in series resonant circuits, so that, instead of being paralleled with a capacitor, they are connected in series between two capacitors. He said that as a result, “We only need to do one cycle,” says Collett. “We step the frequencies; we have no time to ramp. The coin changes the reluctance of the environment.”
This results in a map of applied frequencies in the range 110Hz to 180Hz, versus frequency shifts and changes in amplitude. The whole process is carried out so rapidly, adds Mike Bell, that “a decision to accept or reject is taken in 35 to 40ms, having sorted through up to fourteen parameters.”
The differences between genuine coins and forgeries, and similarly-sized foreign coins, are very small; and along with applying the statistical methods used to distinguish populations of genuine and forged notes, the company also heat soaks the plastic support hardware, so that it is artificially aged to final dimensions and coin acceptors do not go outside calibration. Forgers continue to innovate, as do the makers of validation equipment, so from time to time additional tests have to be incorporated. Forged one-pound coins, in one well established example in wide circulation, are made of white metal, sprayed gold, which reproduces the electrical behaviour of genuine coins. So accepting equipment strikes them against a piezo device, mounted on a special anvil, whose resonant response can be detected as different, should it strike a white metal forgery, instead of the genuine article.


* By examining banknotes at three visible and two infrared frequencies at low resolution, and using novel vision processing software, it is possible to identify banknotes that are either old or have been damaged, and yet distinguish them reliably from forgeries

* Coins can be correctly identified and distinguished from forgeries by their reluctance properties.

* Identification can be undertaken at high speed, using techniques of industrial type prices and ruggedness, and so are equally appropriate for other industrial quality control tasks

This material is protected by MA Business copyright see Terms and Conditions.
One-off usage is permitted but bulk copying is not. For multiple copies contact the sales team.

Related Content


Crane Payment Innovations Ltd

Please view our Terms and Conditions before leaving a comment.

Change the CAPTCHA codeSpeak the CAPTCHA code