- We currently have no target identification technology that can compete with the precision identification capability of human eyeballs1. (DARPATech 2002 Symposium)
- ATR ….Still far from imitating the performance of the eye-brain combination2. (IEEE Transactions on Pattern Analysis and Machine Intelligence
- Vision Systems are Using the Wrong Kind of Information!3 (IEEE 2003 KIMAS Conference)
Biologically inspired sensor technology can enable new tactics and platform concepts for military and commercial applications. In the military domain vision sensors are an essential component of autonomous systems of all types, now receiving rapidly increasing attention (UAV, UGV, ATR etc). However physical vision systems, that substitute for human vision in autonomous systems, have very much less capability for the recognition tasks that must be performed. For example in the DARPATech 2002 Symposium it was noted that "We currently have no target identification technology that can compete with the precision identification capability of human eyeballs1". Manifestly, there is both great need and great potential opportunity for bio-inspired vision sensor technology.
The opportunity is remarkably simple. The human eye derives a very different type and much reduced quantity of visual information4 than physical image sensors. Physical image sensors derive visual information by uniform high density Nyquist sampling, whereas over 99.99% of the retina the human eye derives visual information by non-uniform low- density SD sampling (Figure 1).
Nyquist sampling takes many samples of the spatial-mean, whereas SD sampling takes fewer samples of the spatial-standard deviation of the image within the sample area. In spite of its low sampling density SD sampling thereby captures high-resolution information over the same spatial frequency band as Nyquist. (Each SD sample is a measure of the total high spatial frequency power within the sample area)
The much higher recognition performance of human vision is due, in large part, to the very different type and much reduced quantity of visual information used by human vision. This different (SD) information can easily be derived from existing (conventional) sensor data and then used, as it is in human vision, to much more effectively and more economically perform automatic recognition3,5 and other visual tasks. (Patent Pending)
RPU Technology's technical objective is to transition this different type of information from the biological to the physical domain. Transition of information is very straightforward. Information is information in any domain. If it is effective in one, it can be equally so in the other.
- Richard Wishner, IXO Overview, DARPATech 2002 Symposium
- James A. Ratches, C.P. Walters, Rudolf G. Buser, B.D. Guenther, "Aided and Automatic Target Recognition Based Upon Sensory Inputs From Image Forming Systems" IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol 19, No. 9, September 1997, pp 1004-1019
- Vision Systems are Using the Wrong Kind of Information!. John Merchant, RPU Technology; IEEE International Conference on Integration of Knowledge Intensive Multi-Agent Systems, October 2003, Cambridge, MA
- Sampling Theory for the Human Visual Sense. John Merchant, RPU Technology. Journal of the Optical Society of America, October 1965, Vol. 55, No. 10, pp 1291-1295.
- The Human Visual System Provides Clues to Better Object Recognition. John Merchant, RPU Technology; Photonics Spectra, June 2004 pp102-106.