A new object motion estimation technique for video images, based on a genetic algorithm

This paper describes a new motion estimation technique, that considers object translations, as well as object rotation, zooming and planar multilayering. The proposed approach uses a Genetic Algorithm, that receives the objects of a segmented reference image, and outputs the corresponding motion and laver information, using object and layer genotypes. Genetic operation strategies of reproduction, crossover, mutation, inversion and dominance, are applied recurrently in order to create successive generations of genomes with much better fitness, until convergence, or the maximum allowed number of generations is reached. For the increase of prediction accuracy and convergence speed, life fitness: strategy is used. Simulations with synthetic and natural images have shown very encouraging results with the proposed ME technique, which competes more favorably with respect to the conventional one in accuracy, effectiveness, robustness, simplicity and speed.