In recent years, increasing numbers of military-backed researchers have been borrowing from nature, effectively leveraging millions of years of evolutionary progress. Until we have machines that are as smart, agile and flexible as animals we will have plenty to learn and robots will increasingly come to resemble living things.
The latest trick learned from nature is called "foveal imaging." Here's how I described it for New Scientist:
Inside an animal's eye, an area at the centre of the retina known as the fovea has a higher concentration of light-sensitive cells than surrounding regions. Showing only the centre of a viewpoint in high focus prevents the brain from being overloaded by high-resolution information.
Now, a computer system that mimics this approach using hardware and software is being developed by Nova Sensors, a company based in California, US. It uses a "detection tracking algorithm" to identify windows of interest within a picture, applying tricks such as motion-tracking, tonal analysis and facial recognition.
Unmanned air vehicles need to carry out real-time target detection and tracking, but are limited by bandwidth and processing power restrictions. Thats where foveal imaging offers big benefits: makers Nova Sensors estimate that for a typical 1024x1024 pixel application, their approach reduces the processor throughput by a factor of 15, and cuts the bandwidth requirement by a factor of five.
Just like digital cameras, infrared sensors are offering more and more detailed images from los-res, grainy shots to megapixel and beyond. But while these detailed images are nice, it's kind of a waste to expend processor power to make sure every leaf in a tree comes through clear. Instead, you want detail only where it matters, which is what the foveal system provides. It also has the unique advantage that if another enemy does appear out of the foliage, he will be picked up as an area of interest and a new high-resolution window will form to show him with crystal clarity.
Take the picture above, for instance. The left panel shows output in 'full resolution' mode, like a conventional imager; the right panel shows foveal regions on one and two eyes, shown in different background resolution conditions.
The high-res windows can also following moving targets - The ability to 'fly' numerous foveal windows around the large format FPA [Focal Plane Array] for tracking and surveillance applications is a new capability in the industry, according to Mark Massie, Nova Sensors president.
The system is particularly suitable for tracking moving objects, can produce almost perfectly stabilized images from a moving platform, and recognize pre-defined objects. These are all very handy for UAVs -- moving platforms which spend a lot of time following moving targets and looking for defined objects like trucks or people.
The Air Force is also very interested in foveal sensors for munitions, since they provide new capabilities in for terminal guidance. But ultimately, foveal systems will have much wider applications. All sorts of video applications, from videophones to baby monitoring to home security, will benefit from this technology. I suspect that this will be like the many technologies discussed in Weapons Grade where a military development paves the way for a technology which becomes ubiquitous.
-- David Hambling
© Copyright 2024 Military.com. All rights reserved. This article may not be republished, rebroadcast, rewritten or otherwise distributed without written permission. To reprint or license this article or any content from Military.com, please submit your request here.
