Motion detection and velocity coding of natural scenes by the insect visual
system
David C. O’Carroll1,
Patrick A Shoemaker2, Tamath Rainsford1, Andrew D. Straw1, Eng Leng Mah1 &
Sreeja Rajesh1
1. Discipline of Physiology, School of Molecular and Biomedical Science, &
Centre for Biomedical Engineering, The University of Adelaide, SA 5005,
Australia. 2. Tanner Research, Pasadena, CA.
The insect visual system is a promising model for biomimetic approaches to low-cost computer vision in applications such as guidance systems for unmanned vehicles and collision avoidance hardware. Impressive visual performance is subserved by a surprisingly simple visual system, with a small number of pixels (between 2,000 and 30,000 in a typical insect eye). Insects estimate velocity of targets reliably, yet existing models and silicon implementation of insect motion detectors provide ambiguous, inaccurate estimates of speed. Despite its attractions, the highly non-linear nature of several key stages in insect visual processing presents a challenge to understanding. Our electrophysiological data suggest a variety of non-linear elaborations to existing motion detector models. These include compressive non-linearities and adaptive feedback of local motion detector outputs. Many of these elaborations are relatively easy to implement in biomimetic analog circuits.and may provide greatly improved solutions for velocity analysis and rejection of noise compared with alternative schemes. In addition to discussing our recent advances in neurobiology and modeling this system, I will describe our recent progress in developing hardware based on these models, in both discrete analog circuits and VLSI