Insect Image Processing

Posted 14 Feb 2007 at 03:55 UTC by Rog-a-matic

Insight into the image processing of flying insects might offer a boost to programmers of future airborne robots. Brains of insects are common research subjects. Approximately 100K to 1M neurons process an image of the ground sweeping from front to back across the center of the visual field in order to measure the ratio of the horizontal speed to altitude. Researchers Frank Ruffier and others call the reflex to keep the speed/altitude ratio constant an "optic flow regulator". Relying on this ratio alone eliminates the need to measure speed or altitude. If the insect changes speed, the reflex changes altitude to keep the ratio constant.

Optical flow, posted 14 Feb 2007 at 21:55 UTC by steve » (Master)

I've always found optical flow algorithms fascinating. I think there's a lot of potential there for robotics use.

Another reader sent us a related link to a Scientific American article

Optical flow is the start of other things, posted 15 Feb 2007 at 10:55 UTC by motters » (Master)

Optical flow is closely related to other types of image processing, and its not difficult to imagine how more elaborate forms of visual recognition evolved from it.

Optical flow is fundamentally about tracking the movement of small features over small amounts of time. This is a kind of correspondence problem between successive frames. Correspondence over time and correspondence over space are the same type of algorithm using the same centre/surround neural architecture, so with minor adaptations you can go from optical flow to stereo correspondence with two eyes. The ultimate vision system combines both tracking over time and correlation in space (spatio-temporal correspondence).