Collaborative Research: Recovery of 3D Shapes From Single Views (Purdue University)
Collaborative Research: Recovery of 3D Shapes from Single Views (Temple University)
Researchers at Purdue and Temple Universities have collaborated on the design of a robot that can see natural 3-D scenes just as humans do. The robot forms visual representations of the scene and the objects within it, including the obscured spaces behind objects. This ability is critical for planning navigation in physical environments.
Until very recently, emulating human visual abilities was considered an unsolvable problem. But to interact with humans, intelligent robots must see 3-D scenes and understand what is not directly given by 2-D visual information. The Purdue-Temple team has succeeded in formulating a computational model of human 3-D vision and implementing it in a robot. This brings robots a big step closer to autonomously and safely interacting with humans in such tasks as helping disabled and elderly persons in many of their daily activities, and replacing humans in hazardous situations.
The robot's visual perception relies both on visual data and on prior knowledge about the physical and geometrical regularities that are ubiquitous in our natural environment. Emphasis on prior knowledge, rather than on learning and familiarity, is a fundamental principle of human vision.
Public reaction to latest recession departed from expectationsResearch Areas: People & Society Locations: Indiana, New York
Specialized computer simultaneously analyzes large chunks of informationResearch Areas: Computing Locations: Pennsylvania