Solving Combinatorial Problems with Time Constrains Using Estimation of Distribution Algorithms and Their Application in Video-Tracking Systems [chapter]

Antonio Berlanga, Miguel A., Jess Garca, Jos M.
2010 New Achievements in Evolutionary Computation  
Introduction EDAs (Estimation of Distribution Algorithms) present the suitable features to deal with problems requiring a very efficient search: small populations and a few iterations, compared with the more classic approaches to Evolutionary Algorithms (EAs). The fundamental difference of EDAs with classical EAs is that the formers carry out a search of the probability distribution describing the optimal solutions while EAs directly make the search and provide the solutions to the problem with
more » ... the solutions itself. They share the necessity of codification of solutions by means of binary chains, in the EA terminology they are the "individuals" and the definition of a merit measurement that allows to orient the search direction, the so called "fitness function". In the case of EDAs, operators to manipulate individuals in the search, such as mutation, selections, and crossover, are not needed, since the search is performed directly on the distribution which describes all possible individuals. In this chapter, authors will evaluate the efficiency of EDAs to solve combinatorial problems with time constrains. Specifically, authors will model the visual data association for realtime video tracking problem as a combinatorial problem. Before the application of EDAs to this real-life combinatorial problem, the authors will discuss the application of EDAs algorithms to a classical combinatorial problem, such as the 0/1 knapsack problem, in order to know the complexity degree of the association problem and to find out the most suitable parameters for real-time video tracking problem [1]. The outline of the chapter will be as follows. First, several EDA algorithms will be presented and their evaluation using the theoretical combinatorial problem of 0/1 knapsack problem, which has similar complexity to the association problem in video tracking systems. Next, the mathematical formulation of the Data Association Problem will be shown. Then, the applications of EDA to data association problem, defining the heuristic and the codification, will be presented. Finally, the authors will show the experiments compare the behaviour of several algorithms, taking the advanced Particles-MS tracking as benchmark, in three scenarios taken from two different sources: the publicly available CVBASE [2] and a DV camcorder.
doi:10.5772/8057 fatcat:sye4ie2qwfbpznhopi7iwjk2be