Welcome to the A.L.I.C.E. AI Foundation

Promoting the adoption and development of Alicebot and AIML free software.

Round Sensors &
5000-Pound Brains

Dr. Richard S. Wallace

Dr. Richard S. Wallace

For several years I worked on a visual sensor system called a "logmap" that was based on our understanding of the mapping from the visual field onto the visual cortex. The basis for this mapping was a set of rather cruel animal experiments (cats and monkeys) that involved radiological tagging of active areas in the visual cortex, corresponding to various visual stimuli. My former colleague Eric Schwartz carried out this work in the 1980s.

In the 1990s the U.S. Defense Department gave us money to build various logmap sensors. The basic property of these devices was that the pixel array was arranged in (roughly) concentric circles, with smaller sensors in the center and larger ones in the periphery. We then utilized another invention, the Spherical Pointing Motor, a two-axis direct drive pan-tilt motor, to point the sensor at speeds exceeding that of the human eye (500 degrees/sec). The hypothesized advantage of such a system is that it combines high acuity, random access, and a wide visual field. "Attention" is defined as placing the center or fovea on an object of interest detected in the periphery.

Our logmap sensors were small, 1200-2000 pixels, but the images are quite beautiful and look very natural to our human eye, which is organized the same way. We even developed a method of transmitting the logmaps over voice telephone lines at four frames per second, with analog signaling! With such a low data rate, we could also do real-time visual processing (in 1992) with DSP's and M68332 microcontrollers. We implemented edge detection, motion following, connected components, and some limited face recognition.

The curious thing was that our work was almost completely rejected by the computer vision establishment. They found no use for the round sensors and were (and still are I guess) locked in a Cartesian world of square pixels in square grids. Neural networks people liked it, and so did robotics people, but the vison crowd was most definitely turned off. That was my first experience realizing how influential politics is over science.

Eric Schwartz used to be fond of saying that if the human eye was organized like a TV camera, with the 10giga-pixles you imagine, then we would need a 5000 pound brain to process all the data. If the eye (and visual system) is organized more like a logmap, then you can have your high acuity and wide field of view, but the data rate goes down to something more in TV range. So that's been my working hypothesis for the past decade or so.