Develop two technologies to identify and locate through cameras to make machines smarter

Machines will be smarter when you can see and comprehend things around. This is what researchers at Cambridge University aim to develop two technologies that exploit the ability to learn through camera vision, thereby providing the ability to identify, classify and manipulate objects for the machine. hook.

Two technologies under development are called SegNet and a positioning system has no name. SegNet is a real-time object identification application that can label objects much more accurately than the most advanced radar systems currently used in autonomous cars. SegNet can observe a street view and instantly identify the content in the scene and categorize the content into 12 categories such as roads, road signals, pedestrians, buildings .

The system works in both dark and bright environments and almost in real time . Although currently only tested in an urban environment, SegNet will exploit learning technology to build capabilities and will eventually be able to identify objects in more environments, especially regions. suburbs as well as under various weather and climate conditions.

"SegNet has the ability to identify objects in a photograph very well because it is practiced a lot," says Alex Kendall, a doctoral student at Cambridge University . tweak the system to work better ".

SegNet is taught by graduating students at Cambridge University. They loaded 5000 images with street views, each with pixel areas labeled with classification. SegNet has learned to recognize images over time and eventually does this on its own without the guidance of researchers.

This navigation system can identify learned places with a range of errors in just a few meters.

As my test below, I uploaded a picture of a funny accident . the tank car and SegNet identified the objects in the photo quite accurately. The purple area is the medium, the area is light purple rather than the road, the blue is the sidewalk (actually concrete), the grass color is the trees and the darker than the pedestrian (actually a soldier tank).

Comes with SegNet as a navigation system that works on the same principle. This system helps identify a location based on what is seen through the camera. The system's accuracy is much higher than GPS and works in all situations as long as the camera is visible around, including indoor and outdoor, in tunnels or even under low-light conditions. .

Until now, this navigation system was able to identify learned places with a range of errors in just a few meters. The system works on both map and field levels, such as inside or around a building where the camera is located. In addition, this system can also learn about the location around the location used. Developers believe that the immediate system will be used on indoor robots and will eventually expand to many other mobile objects such as unmanned cars or wearable devices.