Renato Salas-Moreno and colleagues have added object recognition to a computer vision technique called simultaneous location and mapping (SLAM).
Renato Salas-Moreno a PhD student here in the Department of Computing supervised by Professors Andrew Davison and Paul Kelly has been featured in a New scientist artice entitiled "Remembering objects lets computers learn like a child". The article talks about Renato's and his collegues work in the Robot Vision Group on SLAM. Renato's research is supported by a PhD scholarship from AMD
The Robot Vision Group conducts research on real-time computer vision techniques applicable to robotics or other demanding real-world, real-time applications. Their particular focus is on problems related to visual SLAM (Simultaneous Localisation and Mapping).
SLAM is a process by which a mobile robot can build a map of an environment and at the same time use this map to deduce it's location.
In a new system, called SLAM++, the computer constantly tries to match the points and lines it sees to objects in its database. As soon as it finds a shape it can identify – often after seeing only a part of it – that area of the map can be filled in. Currently, the database is prepared by hand, but the next version will allow the system to add new objects itself as it encounters them. "It's similar to how a child learns about the world," says Salas-Moreno.
The database also lists the properties of the stored objects. So when the computer recognises a chair, it will know what they are used for, how much they typically weigh, and which way up they go. This knowledge will help digital avatars interact with the real world in augmented reality applications.
The SLAM++ system was demonstrated live to the general public at the Imperial Festival in May this year.
To read the full article see http://www.newscientist.com/article/mg21829205.800-remembering-objects-lets-computers-learn-like-a-child.html
Article text (excluding photos or graphics) available under an Attribution-NonCommercial-ShareAlike Creative Commons license.
Photos and graphics subject to third party copyright used with permission or © Imperial College London.