Researchers help robots see better

A new algorithm by which robots can recognize a randomly oriented object has been developed by Jared Glover, a graduate student in MIT's Department of Electrical Engineering and Computer Science, and Sanja Popovic, an MIT grad who is now at Google. Based on Bingham distribution, a statistical technique often used in analysis of data about historical changes in Earth's magnetic field, the algorithm is 15% better than its best competitor at identifying familiar objects in cluttered scenes. What's more, in cases where visual information is particularly poor, the algorithm offers an improvement of more than 50% over the best alternatives, achieving significantly more reliable object detections.

The researchers say their main contribution is a new method called Bingham Procrustean Alignment (BPA) to align models of objects within a 3-D image of a scene. BPA uses point correspondences between oriented features to derive a probability distribution over possible model poses. The orientation component of this distribution, conditioned on the position, is shown to be a Bingham distribution. For further information, contact Jared Glover at .

The cluttered image at left was analyzed using the new Bingham Procrustean Alignment algorithm. The models of the objects as detected are shown at left, along with the feature correspondences that BPA used to align the model. Surface features are indicated by red points, with lines sticking out of them to indicate orientations (red for normals, orange for principal curvatures). Edge features (which are orientation-less) are shown by magenta points.