Robots Dont See as Well as You Do
Robot designers are still working on ways to emulate the human eye. Just when you thought digital cameras were all the rage, we learn from EurekAlert they are miserable substitutes when put into the eye sockets of robots. Robot-vision export Vladimir Brajovic explains:
Often, when we take a picture with a digital or film camera, we are disappointed that many details we remember seeing appear in the image buried in deep shadows or washed out in overexposed regions. This is because our eyes have a built-in mechanism to adapt to local illumination conditions, while our cameras don’t. Because of this camera deficiency, robot vision often fails. (Emphasis added.)
But can’t automatic exposure meters do the same thing? No; they pick either a spot or average of the scene, and adjust all the sensors to the same level. Our individual rods and cones not only have individual light adjustments, but “talk to each other” about what they see, and do image processing before the signals reach the brain (see 12/30/2003 and 05/27/2003 and 05/22/2003 entries). Brajovic is trying to develop image sensors with some of these desirable capabilities.
Werner Gitt provides many more gee-whiz statistics about the eye, and other body senses, in a wonderful book, The Wonder of Man. One square millimeter of retina has 400,000 sensors. The photoreceptors are so sensitive, a single photon can activate them. The rods can react in 0.3 seconds, the cones in 0.075 seconds. Three types of cones, sensitive to different wavelengths, give us complete coverage of the visible light spectrum, with over 300 discernible hues. Unlike film, which is rated for a particular “speed” or sensitivity, the eye’s photoreceptors are sensitive over 5 powers of 10, or 100,000 to one. The signals are transmitted on two separate channels then recombined, to avoid the problem of thermal noise. The optic nerve also filters out noise by sensing the response from multiple rods within a time limit of 0.02 second, and sending the signal along only if there is nearly simultaneous response from four or five rods scattered across the field.
Rhodopsin, the light-sensing protein, is a large molecule composed of 350 amino acids (see online book). To avoid saturation of the photoreceptors, muscles move the eyes in constant jerks called saccade’s, but the optic nerve and brain automatically compensates for the motion (see 03/29/2002 entry). The optic nerves cross behind the eyes, and are received by opposite sides of the brain; there, the inverted images are flipped and recombined seamlessly. The brain even fills in the blanks caused by the holes where the optic nerves leave the retina, by interpolating these “blind spots” with similar pixels from the surrounding field. Small as they are, the eyes have 120 megapixel resolution and can separate angles of one minute (1′) of arc. What’s more, they can do it in full-motion stereo. With the help of the iris, eyelid, retina and other mechanisms, we can see everything from dim stars to bright sunlight reflected off snow: a phenomenal range of sensitivity – a factor of 1 million million to one. And our eyes are not even the sharpest or most sensitive in the animal kingdom. Need more? Buy Gitt’s book. After you enjoy it, hand it to an evolutionist. Perhaps, though, these days, intentionally giving someone cold shudders would be misconstrued as cruelty (see 07/13/2001 commentary).