Voices

You can be identified by how you drive

27 May 2016

Robert Pepper 1

While racing car teams always knew it, a new report shows that your style of driving can identify you.

MORSE CODE is a simple code that represents every letter of the alphabet as a combination of dots and dashes.  For example, S is … and O is — so SOS (Save Our Souls, the international emergency signal of distress) is …—…—… repeated.

When Morse was widely used there were many highly skilled operators who could rapidly transmit messages, and while each would tap out the same sequence of dots and dashes, the way they did so varied ever so slightly. The fractional delay between dots or dashes, or letters, and the length or a dot or a dash were all minutely different between operators. These differences added up to a unique ‘fist’, discernible to the trained ear, such that each operator could be distinguished by their fist alone. In the same way, if you asked two chefs to prepare the same meal enough times there would be slight variations unique to each chef that allowed the diner to work out who made the meal, and the manner of using a computer keyboard is also unique to each individual.

So it’s perhaps not too surprising that the way we drive a car is also unique to each individual, and that it appears possible to identify a given driver with a very high degree of certainty just from the way they drive. Race car teams already know this as they’re able to identify which of their drivers drove a lap by looking at the test data – and this is also why two identical racecars are set up differently for the same driver – and now researchers at the University of Washington has just proven the point for everyday drivers with a series of tests.

The researchers selected 15 drivers, 8 male, 7 female, ranging in age from 24 to 47, and had them drive exactly the same car through a closed-circuit course and then an open road course of 80 km. The same time of day and even the same music was selected for the drive, and the drivers were familarised with the car before they began the test.

As they drove, the car collected data from 16 sensors covering the likes of brakes, steering, accelerator, gears, engine RPM, speed and yaw, both when the control was used and how. This information was collected via the OBD port, a now-standard interface on all cars that allows easy collection of diagnostic data.

The results were startling. According to the team, it was possible to identify with 100% certainty which driver was which by analysing the data. In fact, use of the brake pedal alone was sufficient to identify drivers after driving the entire course, and after only about 15 minutes with the brake pedal data alone there was an 87% chance of identifying the driver. From the report:

The maximum engine torque is the amount of torque the engine can produce at any given situation, different from the actually requested torque.

That is quite amazingly accurate, and it is also worth noting that only primary car controls were monitored. If ancillary controls such as the angle of the steering wheel, seats, mirrors, use of heating/cooling, preferred music and volume were considered then the speed of identification would be even quicker. Even the method of operating the car could be a factor; for example, in most cars there are two ways to change the music volume, and several ways to display any given information on a screen. The pre-start routine would be unique too; do you put the seatbelt on and then start, or vice-versa, how hard you pull the parkbrake on…the potential ways to identify an individual are vast.

There are a few interesting points about the test. First, a sample of 15 is very small, and they were careful to choose different gender and ages. That may be because they didn’t expect to find much difference even with widely different driver types. Aside from a bigger sample, it would be have better to have selected drivers as similar as possible; eg. all males between 20 and 25 with similar levels of experience, or all females 40-50. The differences in those groups should have been smaller, and better able to test the algorithms used to identify drivers.

Yet even with a small sample of 15, the wider experiences with something as simple as Morse code indicate that our method of driving – a complex operation – can be used to identify individuals.

So what are the implications if a driver can be identified by their driving technique? Or, if it can be known that a specific driver is NOT driving the car?

The researchers explored a few possibilities. One is that banned drivers are identified, and cars simply refuse to operate once they realise they are being driven by a banned driver. Of if you lend a car to a friend you can see if anyone other than that friend drives it. Same deal for rental companies, they could not only see where their car is, but identify how many drivers have driven it and if they can match the data, then who those drivers were.

Another is that low-risk behaviours by low-risk drivers are modeled, and insurance discounts given to those drivers who match low-risk behaviours. Or, pre-empetivley and more controversially, drivers displaying high-risk behaviour are penalised even before a crash. Driver and situation-based insurance has been tried and in some cases is working, but it’s definitely a model of charging that will become more prevalent over time.

A spooky possibility is that everyone has their driving behaviour modeled, and every car reports back, in real time, who is driving it. Then when the police need to catch a criminal they need only wait till he or she drives a car before they know where to go to catch their felon. Combine that with all the data the likes of Facebook, Twitter, Apple, the banks and everyone else is collecting and I think we can say that privacy is well and truly dead.

Could you ‘fake’ your style? That may seem easy but in reality is very hard to do. Again we need to look at sophisticated pattern recognition for voices, handwriting and the like which can detect attempts to fool the system. Regular readers of Practical Motoring will be able to tell whether myself, Isaac or Paul has written an article without needing to see the byline, and I couldn’t possibly copy Isaac’s style……..

Overall, the possibilities are exciting but also frightening, and all the more so because at this early stage history tells us we cannot possibly foresee all the effects and side effects. Nevertheless, the technical advances and data collection are not going to stop any time soon.

Source report: http://www.autosec.org/pubs/fingerprint.pdf