From hough to darwin: An individual evolutionary strategy applied to artificial vision

Abstract

This paper presents an individual evolutionary Strategy devised for fast image analysis applications. The example problem chosen is obstacle detection using a pair of cameras. The algorithm evolves a population of threedimensional points (‘flies’) in the cameras fields of view, using a low complexity fitness function giving highest values to flies likely to be on the surfaces of 3-D obstacles. The algorithm uses classical sharing, mutation and crossover operators. The final result is a fraction of the population rather than a single individual. Some test results are presented and potential extensions to real-time image sequence processing, mobile objects tracking and mobile robotics are discussed.