Almost all the successful inventions in the history of science are inspired by some masterpiece of nature. Similarly the close observation of vision system underlying a fly reviled some extraordinary but mysterious mathematical non-linear equations, which will help in mastering the vision system of future battlefield drones, search-and-rescue robots and other miniaturized systems, where computation power is the prime issue. These mathematical equations provide and ultra-efficient method to judge future motion pattern from a stream of visual data. As a matter of fact, although the research team form Australia’s University of Adelaide built this system, the team is unable to understand its working. According to David O’Carroll, a computational neuroscientist form the same university said “The number of computations involved is quite small. We can get an answer using tens of thousands of times less floating-point computations than in traditional ways.”

The algorithm developed based on these mathematical equations computes data from camera in a series of five stages. Each stage implements a separate equation and these equations represent tricks used in vision system of fly to handle different levels of contrast, brightness and motion. The parameters to these equations keep on changing with inputs. Instead of computing frame-by-frame comparison of last pixels, the algorithm works more like a video compression system and emphasizes on large scale patterns of change. The algorithm was tested using a high resolution animation and a program from operating robot. Results reviled that the system although vary in ways a usual baffle motion detector but worked in a range of natural lighting conditions.

Before this the best known method for detecting motion changes is named Lucas-Kanade method, it calculates up-down, side-to-side motion changes by studying changes every pixel undergoes in a visual filed, on frame by frame basis. It is in use in most of experimental UAVs visual systems, because of its high power consumption and bruit-force approach, is not suitable for smaller systems. These new algorithm inspired by lowly fly is best suited for small flying robots.

The research team got this algorithm by studying the neural circuits, behind visual system of a fly, only related to side-to-side yaw. Following the similar techniques other optical flows, like ones produced during forward and backward motion in 3D space, can be computed. This will ultimately not only help in future battlefields and search and rescue operations but will also improve other areas related to  optical vision of miniaturized systems.

Reference: “Robust Models for Optic Flow Coding in Natural Scenes Inspired by Insect Biology.” By Russell S. A. Brinkworth, David C. O’Carroll. PLoS Computation Biology, November 6, 2009.