Dynamic flies: a new pattern recognition tool applied to stereo sequence processing

The "fly algorithm" is a fast artificial evolution-based technique devised for the exploration of parameter space in pattern recognition applications. In the application described, we evolve a population which constitutes a particle-based three-dimensional representation of the scene. Each individual represents a three-dimensional point in the scene and may be fitted with optional velocity parameters. Evolution is controlled by a fitness function which contains all pixel-level calculations, and uses classical evolutionary operators (sharing, mutation, crossover). The combined individual approach and low complexity fitness function allow fast processing. Test results and an application to mobile robotics are presented. (C) 2002 Elsevier Science B.V. All rights reserved.