A bio-inspired visual collision detection mechanism for cars: Optimisation of a model of a locust neuron to a novel environment

Authors: Yue, S., Rind, F.C., Keil, M.S., Cuadri, J. and Stafford, R.

Journal: Neurocomputing

Volume: 69

Issue: 13-15

Pages: 1591-1598

ISSN: 0925-2312

DOI: 10.1016/j.neucom.2005.06.017

Abstract:

The lobula giant movement detector (LGMD) neuron of locusts has been shown to preferentially respond to objects approaching the eye of a locust on a direct collision course. Computer simulations of the neuron have been developed and have demonstrated the ability of mobile robots, interfaced with a simulated LGMD model, to avoid collisions. In this study, a model of the LGMD neuron is presented and the functional parameters of the model identified. Models with different parameters were presented with a range of automotive video sequences, including collisions with cars. The parameters were optimised to respond correctly to the video sequences using a range of genetic algorithms (GAs). The model evolved most rapidly using GAs with high clone rates into a form suitable for detecting collisions with cars and not producing false collision alerts to most non-collision scenes. © 2005 Elsevier B.V. All rights reserved.

Source: Scopus