Forget GPS, here are smartphone based cars powered by SegNet

Researchers have recently developed two new smartphone-based systems powered by SegNet, which can accelerate the development of driver-less cars by identifying a user’s orientation and location in the places where GPS does not operate.

These systems can also identify various components of a road scene on a regular camera or smartphone, performing the same work as sensors which cost millions. 

Professor Roberto Cipolla from University of Cambridge has led the research. He said that the vision is our most powerful sense and this sense needs to be acquired by these machines. But, teaching a machine to see is more difficult than it sounds. Currently the system cannot control a driverless car. However, the ability to provide a machine with a vision and accurately identify what it is looking for and where it is, is a vital part of developing autonomous robotics and vehicles.

What is SegNet?

SegNet is the first system, which can take an image of a street scene it hasn’t visualized before and classify it by sorting the objects into twelve different categories such as street signs, roads, buildings, cyclists and buildings in real time.

Alex Kendall, a PhD student added that it is remarkably good at recognizing things in an image because it has had so much practice.

It can deal with shadow, light and night- time environments and is currently labeling more than 90 percent of pixels correctly. SegNet was primarily trained in urban and highway environments so it has some learning to do for snowy, rural or desert environments – although it has performed well in initial tests for these environments.

There are three important technological questions that must be answered to design autonomous vehicles what do I do next, where am I and what’s around me. SegNet addresses the third question while a separate but complementary system answers the second by using images to determine both orientation and precise location.

The authors noted that the users can visit the SegNet website and upload an image or search for any town or city in the world and the system will label all the components of the road scene. The system has been successfully tested on both city motorways and roads.

The second localization system runs on a similar architecture to SegNet and is able to determine and localize their orientation from a single color image in a busy urban scene.

The localization system uses the geometry of a scene to learn its exact location and is able to determine; for example, whether it is looking at the west or east side of a building, even if the two sides appear identical.

Better than GPS

The system is more accurate than GPS and can operate in places where GPS can’t, such as in tunnels, indoors or in cities where a reliable GPS signal is not available. The researchers presented details of these technologies at the International Conference on Computer Vision in Santiago, Chile.

Cipolla also added that in the short term we are more likely to see this type of a system on a domestic robot such as a robotic vacuum cleaner for example.

Source: NDTV

Image Sources: Image 1 & Featured Image,Image 2.