I watched a show on Tuesday evening. The kind of show that I really love in a nerdy way. Not one about sex, or drugs, or violence, or murder mystery (although I do enjoy those as well), but, a real, proper, honest-to-god BBC2 documentary (please note, other deities are also available).
For those poor souls who dwell outside the UK who haven't yet experienced the pure joy of one of these series, it's the equivalent of a nice warm blanket, accompanied by a roaring fire and a cup of coca, on a miserable Irish winters night, with a proper Atlantic gale howling outside.
The topics of these documentaries are varied and too numerous to mention here. I do remember a very decent one last year about how the water system in the UK functions, and there are always great ones about this and that - the banalities of modern English life. And so this show in question went - all about the lovely folk who look after the motorways in England: The Motorway: Life in the Fast Lane
We learned about the M6, the folk who arrive first at crash scenes, those who talk people down from bridges, and the lovely people who watch the traffic behind banks of monitors, linked-up to hundreds and thousands of CCTV cameras.
In amongst all of this awesomeness there was a segment that caught my attention (around the 12 minute mark if you have access to BBC iPlayer). We were introduced to two chaps: Steve and Jed. Charming fellas who described their role, in the wonderful orchestrated process of managing the motorways, as spending 5 hrs a day, 3 days a week, driving the motorways at 50 mph looking for potholes. One of the chaps did the driving - avoiding potholes and other traffic I assume - and the other was the lookout.
If Steve, the lookout, spotted a pothole, he'd press a neat little button on the survey tool provided on the dashboard that logged the location for the maintenance crew to follow up later. For larger potholes, that needed immediate attention, the spotter could place call to home base for an immediate fix and another team would roll out, stop traffic and get it sorted immediately.
So my little data science mind got whirring.... Surely there is a more efficient way than having two chaps driving the motorways of the UK looking for holes in the ground? Couldn't the wonders of science, big data, machine learning, and perhaps the greatest thing on the Gartner hype cycle: the Internet of Things (IoT), come together in beautiful scientific and software symmetry to solve this glaring inefficiency?
Today in the office, at the end of a long day of data-mining, I broached the thorny pothole observation issue with my wonderful colleagues.
My initial thought was do away with the car altogether and perhaps use some image capture and recognition to find potholes. I mean, if the Irishman who has an IQ greater than Einstein could use image recognition to find the Boston Bombers, no reason we couldn't use a similar algorithm applied to potholes, right? We'd need to spend some time thinking about the right training set and which image recognition patterns would derive the best results, taking into account: weather, traffic flows, varying light contrasts, and a image quality. Might be a bit tricky, but, I'm sure we could hack something together. We could ask the nice folk at Inmarsat based here at Old St roundabout to help a local startup and redirect a few of their passing commercial satellites to scan the universe of possible motorways in the UK. Then we could grab a few thousand AWS instances and distribute the processing job, stick an awesome visualisation on the front in real time and and Bob's your father's brother.
Unfortunately, under scrutiny, this idea was exposed as a little silly. Redirecting those satellites would cost a bloody fortune, we don't really know anybody at Inmarsat that we could ask, and it might slow down my TV signal or, worst still, my phone reception.
So onto plan B (fail fast right?) They spent a lot of the show showing all the fancy CCTV cameras that cover every inch of the motorways. OK, the resolution isn't going to be as good as a commercial satellite, however, you just had to build an additional margin for error into the algorithm. You'll probably get a few more false positives than is necessary, but, the downside there isn't so bad - arriving at a piece of tarmac that didn't actually need repairing.
Then James chipped in with an orthogonal view on the whole issue. Just stick a frickin'-laser-beam on the front of the car that measures the depth of the road ahead. You could have some kind of expectation function for the distance to target and from there just look for the outliers, presumably where the distance is greater than expected. If it's less than expected I'd imagine you are about to crash. Problem solved, right? You still need somebody driving the car (at least until Google gets around to fixing the whole driverless car thingy) but you can save a bundle on the spotter. Sorry Steve, you are reassigned to new duties.
But then this was dismissed as being cruel. Jed, the driver, would be left sitting there on his own the whole time with nobody to chat to about the weather, Scottish elections, David Cameron, or the football, etc. If memory serves, our new data scientist Will suggested adding a Siri-like semantic voice system to the car. This could interact with Jed on topics of his choosing, perhaps hooked up to Yahoo's recent acquisition, Storyful to deliver relevant news content. Or better still, the voice-system, yet to be named, could be trained to pick up on local areas of interest adjacent to the motorway by geo-location, which could be read out to prevent loneliness or boredom. Ok, slight project creep into the world of emergent semantic technologies, but, what the hey.
Lastly we came up with a glorious IoT solution. Most IoT articles your read imagine kettles, toasters and ovens talking to smart-phones. I've never quite understood what they would actually say to each other. "God damn it toaster, you burnt the crumpets again!' or words to that effect, although in binary. Anyhow I digress, our IoT solution relies on sensors, GPS, and the added bonus of a glorious distributed computing question.
Mount sensors on the suspensions of all web-enabled motorcars (marketing folk will call cars that eventually, even if they aren't already). These sensors will detect when the car runs over a sufficiently impressive divot in the pavement. This will trigger a messaging system that fires the on-board GPS to capture the location of the deviation. This message is then communicated to a base station, perhaps located at the nearest electric charging services. These events are then aggregated across multiple vehicle reports. Once a certain density score is hit, then you've got a confirmed pothole and the jolly repair crew can be mobilised.
The problem with this is that there aren't really enough web-enabled motors on the road. Much as I'm sure the folk at the auto-makers would be delighted if we all rushed out and purchased new cars, the reality is that reaching sufficient monitoring density is going to take a few years more. So that probably rules out the IoT solution for now.
We therefore logically concluded that the existing solution is good enough and our two hero's Steve and Jed are safe in their jobs for now. Data science and the IoT haven't conspired to steal their jobs in the smooth functioning of the motorways just yet.
Long live the humble pothole.