In computer vision, robots typically identify objects in their field of view using a data set. From a collection of dozens or even thousands of images, a robot can be taught to identify specific markers to improve its estimation of the desired object (be it a face, car, whatever). The problem is in cluttered or poorly lit images (the kind one would expect to encounter in daily life), objects can be difficult to make out from the background. That’s where humans come in. Humans can draw polygons around objects and give the object a name to help the robot out. Alex Sorokin has developed an application that will allow humans to do just that, by harnessing the available workforce on Amazon’s Mechanical Turk. Perhaps a new kind of reCaptcha could be made as well. Video explanation follows after the break.
[source: Willow Garage]