On the Evolution of Formal Models and Artificial Neural Architectures for Visual Motion Detection

Abstract

Motion is a key, basic descriptor of our visual experience of the outside world. The lack of motion perception is a devastating illness that leads to death in animals and seriously impaired behavior in humans. Thus, the study of biological basis of motion detection and analysis and the modelling and artificial implementation of those mechanisms has been a fruitful path of science in the last 60 years. Along this paper, the authors make a review of the main models of motion perception that have emerged since the decade of the 60’s stress-ing the underlying biological concepts that have inspired most of them and the traditional architectural concepts imprinted in their functionality and design: formal mathematical analysis, strict geometric patterns of neuron-like processors, selectivity of stimulate etc. Traditional approaches are, then, questioned to include “messy” characteristics of real biological systems such as random distribution of neuron-like processors, non homogeneity of neural architecture, sudden failure of processing units and, in general, non deterministic behavior of the system. As a result is interesting to show that reliability of motion analysis, computational cost and extraction of pure geometrical visual descriptors (size and position of moving objects) besides motion are improved in an implemented model.