A team of researchers from Cambridge University is borrowing some of the techniques used in autonomous vehicles to teach your phone to navigate, even when it doesn’t have access to positioning information like a GPS signal.
In fact, the team has developed two new pieces of software that run on mobile phones but think a little like driverless cars. The first, called SegNet, can take footage of a street scene from a smartphone’s camera and classify it, sorting objects into 12 different categories – such as roads, street signs, pedestrians, buildings and cyclists.
You can see it in action in the GIF above. The team explains how it works:
SegNet learns by example – it was ‘trained’ by an industrious group of Cambridge undergraduate students, who manually labelled every pixel in each of 5000 images, with each image taking about 30 minutes to complete. Once the labelling was finished, the researchers then took two days to ‘train’ the system before it was put into action.
The team claims “it can deal with light, shadow and night-time environments, and currently labels more than 90% of pixels correctly.”
The second system uses footage from the street scene to discern where the phone is, using the geometry of buildings and street furniture to discern location. The system has so far been tested in the centre of Cambridge, where it’s been shown to be “far more accurate than GPS”. It’s very similar to the way most autonomous cars, including those made by Google, compare sensed data to existing prior maps to orientate themselves on the road.
Don’t think that your phone will suddenly turn your car into an autonomous vehicle, though. “In the short term, we’re more likely to see this sort of system on a domestic robot – such as a robotic vacuum cleaner, for instance,” explained Roberto Cipolla, one of the researchers working on the project, in a press release. Gotta start somewhere, even if it sucks, eh? [Cambridge University]