Man-made velocity estimators based on insect vision

Abstract

The study of insect vision is of significant interest to engineers for inspiring the design offuture motion-sensitive smart sensor devices, for collision avoidance applications. Althoughinsects are relatively simple organisms compared to vertebrates, they are blessed with avery efficient visual system, which enables them to navigate with great ease andaccuracy. Biologically inspired motion detection models are bound to replace theconventional machine vision technology because of their simplicity and significantadvantages in a number of applications. The dominant model for insect motiondetection, first proposed by Hassentein and Reichardt in 1956, has gained widespreadacceptance in the invertebrate vision community. The template model is anotherknown model proposed later by Horridge in 1990, which permits simple trackingtechniques and lends itself easily to both hardware and software. In this paper, wecompare these two different motion detecting strategies. It was found from the dataobtained from the intracellular recordings of the steady-state responses of wide-fieldneurons in the hoverfly Volucella, that the shape of the curves obtained agreewith the theoretical predictions made by Dror. In order to compare this with thetemplate model, we carried out an experiment to obtain the velocity responsecurves of the template model to the same image statistics. The results lead us tobelieve that the fly motion detector emulates a modified Reichardt correlator.In the second part of the paper, modifications are made to the Reichardt detector thatimprove its performance in velocity detection by reducing its dependance on contrast andimage structure. Our recent neurobiological experiments suggest that adaptive mechanismsdecrease the EMD (elementary motion detector) dependence on pattern contrast andimprove reliability. So appropriate modelling of an adaptive feedback mechanism is carriedout to normalize contrast of input signals in order to improve the reliability and robustnessof velocity estimation.