Multiobjective tuning of a multitarget tracking algorithm using an evolutionary algorithm

Multitarget tracking MTT algorithms have been tuned by a variety of optimization methods using a single objective, but only recently have they been tuned with multi-objectives technique. The desire to compare single-objective MTT algorithms using numerous metrics is well documented in the literature for over a decade. We discuss an experiment to quantify the need or lack of need for Monte Carlo (MC) runs in tuning the parameters of a MTT algorithm using some of these metrics. The extreme computational requirement of running a MTT MC experiment for each individual evaluation function drives the need to determine the worth of doing so. The results of using a single run are compared to that of using a MC evaluation with multiple runs as compared to a multiobjective evolutionary algorithm approach. Additional analysis is performed on the search space demonstrating other useful information the decision maker may use to select an optimal operating point from a calculated Pareto front.