Sample Photographs

The vision system is very dependent on lighting conditions, but this is a problem for any photographer with either a traditional or a digital camera. In the photos below, you'll see a variety of lighting conditions, poses, and environments. The 'robot vision' images show detected skin in blue, detected faces with a blue box, and laser rangefinder data as a graph a the bottom; close objects are shorter lines.

Over 3,000 photos were taken at SIGGRAPH2002 alone, but due to privacy issues we cannot post all of them. If you would like additional photos, please contact us.

Photos of one person are centered, with the head positioned in the upper 1/3 division. Lighting in this shot is a mixture of fluorescent and natural.
[robot vision]
Pictured: Zach Byars

This shot illustrates a combination of foreground and background objects, as well as posed and candid shots. It is difficult if not impossible for a robot to identify these traits, but does it matter? Out of all the photos taken by Lewis, some will be good, some will be bad, much like a traditional photographer. Lighting is a mixture of fluorescent and natural.
[robot vision]
Pictured, left to right: Rennie Bailey, John Larson, Patrick Vaillancourt, Ben Westover, Adam Mitz, Kevin Goodier

Lewis attempts to closely frame shots of multiple people without cropping anyone out of the picture. The face finder responds equally with frontal or portrait views. This image is from SIGGRAPH2002 and shows the difficult lighting conditions. The photo area was essentially lit by four 1000W spotlights positioned approximately 10 feet off the ground at each corner. Despite the diffusing screens, the lights were still blinding.
[robot vision] (note: slightly different from the others, only shows skin in green)
Pictured, left to right: Dr. Cindy Grimm, Jacob Cynamon

Posed shots work equally well, as shown in this photo. When the photo area is less crowded and people can pose and interact easily with the robot, fewer candid shots are taken. You can see how Lewis zoomed the camera in to fill the frame. This shot was also taken at SIGGRAPH2002 and the harsh lighting is clearly visible in the high regions of contrast on the faces.
[robot vision]
Pictured, left to right: Sean Fitzroy, unknown

Again, despite poor lighting condiitions, the face finding algorithm detected both light and dark skin. The middle person is behind the other two, but the laser range-finder still found the person and the face. The three people are evenly spaced in the photo.
View [robot vision]
Pictured: Roz Dimon, RDA Design; Michael Dixon; Rennie Bailey