Evolutionary photogrammetry using volumetric representation

This thesis is an examination of the use of a Genetic Algorithm (GA) to evolve a volumetric representation of a 3D object from multiple images. The chromosomes create a finite grid of floating point values at regular intervals which represent the amount of occupation at each of the points of intersection. A mutation and crossover operator are developed and their variations analyzed with respect to the amount of benefit to the algorithm. The fitness of each chromosome is determined by rendering an iso-surface created using the Marching Cubes algorithm from each of the image perspectives. The renderings are then subjected to pixel by pixel comparisons with the original images and the differences are summed to give a fitness value. Methods for enhancing the performance of the algorithm such as reduction in the number of grid points and various settings of the operators are discussed.