| dc.description.abstract | Over recent years, Evolutionary Algorithms (EA) have emerged as a practical approach for solving hard optimization problems ubiquitously presented in real life. The inherent advantage of EA over other types of numerical optimization methods lies in the fact that they require much less or no prior knowledge of the objective function. Differential Evolution (DE) has emerged as a highly competitive and powerful real parameter optimizer in the diverse community of evolutionary algorithms. The study of this dissertation is focused on two main approaches. The first approach focuses on studying and improving DE by creating its variants that aim at altering/adapting its control parameters and mutation strategies during the course of the search. The performance of DE depends largely upon the mutation strategy used, its control parameters namely the scale factor F, the crossover rate Cr, and the population size NP, and is quite sensitive to their appropriate settings. A simple and effective technique that alters F in stages, first through random perturbations and then through the application of an annealing schedule, is proposed. After that, the impact and efficacy of adapting mutation strategies with or without adapting the control parameters is investigated. The second approach is concerned with the application side of DE which is used as an optimizer either as the primary algorithm or as a surrogate to improve the performance of the overall system. The focus area is video based vehicle classification. A DE based vehicle classification system is proposed. The system in its essence, aims to classify a vehicle, based on the number of circles (axles) in an image using Hough Transform which is a popular parameter based feature detection method. Differential Evolution (DE) is coupled with Hough Transform to improve the overall accuracy of the classification system. DE is further employed as an optimizer in an extension of the previous vehicle detector and classifier. This system has a novel appearance based model utilizing pixel color information and is capable of classifying multi-lane moving vehicles into seven different classes. Five different variants of DE on varied videos are tested, and a performance profile of all the variants is provided. | en_US |
