A team working at Brown University, believe they have made a breakthrough in robotic vision systems: A robot that follows a person from room to room to room, tracking them regardless of lighting conditions, crowds milling between the user and the robot, temporary obstructions or anything.

This extraordinary-seeming development isn't as big a leap as it might initially seem however. The researchers have simply taken machine vision from 2D to 3D.

The team developed a camera designed specifically for depth-imaging. It uses not one, not two but an array of dozens of lenses to allow it to pinpoint in real-time the depth of field of any given object - like say a person. If another, larger person walks in front of them, the robot is not distracted because it can tell the new person is far too close to the camera - they are not at the focal depth if you will.

The camera uses infrared, allowing the robot to identify and focus on the silhouette of a user, largely irrespective of lighting conditions. Because it is not using the visible light spectrum, it is unphased by changes in the visible light spectrum, by and large. Outdoor work still confuses it however.

Finally, unique image recognition algorithms have been developed that allow the robot to accurately interpret the images returned by the depth perception camera.

"This allows us to reasonably segment people from background scenery with less dependence on lighting condition," said Chad Jenkins, an assistant professor of computer science at Brown University who presented the work at the 2009 Human-Robot Interaction conference in California, US.

References

Wag the Robot? Brown Scientists Build Robot That Responds to Human Gestures

Robot Plays Follows the Leader