Communications of the ACM
Better Robot Vision
A newly developed robot-vision algorithm is 15 percent better than its best competitor in identifying objects in cluttered scenes.
Credit: Robodoki
Massachusetts Institute of Technology (MIT) researchers using the Bingham distribution have developed a robot-vision algorithm they say is 15 percent better than its best competitor at identifying familiar objects in cluttered scenes.
The researchers, led by MIT graduate student Jared Glover, are using Bingham distributions to analyze the orientation of ping pong balls in flight, in an attempt to teach robots to play ping pong. Glover also has developed a suite of software tools that accelerate calculations involving Bingham distributions.
Glover shows that the rotation probabilities for any given pair of points can be described as a Bingham distribution, which means they can be combined into a single, cumulative Bingham distribution. Glover's research enables the algorithm to explore possible rotations in a principled way, quickly converging on the one that provides the best fit between points.
The researchers believe that additional sources of information could improve the algorithm's performance even further. "It's a better representation, so I think once it's understood, this'll just kind of become one of the things that is built in when you're doing the 3D fits," says Magic Leap's Gary Bradski.
From MIT News
View Full Article
Abstracts Copyright © 2013 Information Inc., Bethesda, Maryland, USA
No entries found