Evaluation of association algorithms and hypotheses combination

In this dissertation we study several aspects of the data association problems arising in multisensor-multitarget tracking. ^ The first part of this dissertation deals with the design of a track association and fusion approach for a networked surveillance system. A hierarchical processing of the ambiguous track data produced by the communication system is proposed. Two methods to obtain the association cost are compared and three association algorithms are also compared in order to characterize their performance both in terms of the quality of the association as well as the computational load. Among the contributions of the first part is the evaluation of the sequential m-best 2D algorithm on a realistic problem and the development of a mechanism by which the Lagrangean relaxation based algorithm does provide complete solutions. The second part of the thesis describes a method for obtaining, in closed form, the performance of data association in terms of the misassociation probability for scenarios with 2 targets. The cases of measurement-to-track and track-to-track association are considered. For the first case, the exact probability of misassociation is obtained for the cases of equal and nonequal innovation co-variances. The latter case is much more complex, so approximations are used to evaluate the desired probabilities. Monte Carlo simulations are performed, and they confirm the validity of the results. ^ The third part deals with the combination of the top m hypotheses obtained using a Multiple Hypotheses Tracker. The idea behind such a combination is that the set of hypotheses with cost near the optimum contain more information than just the single top hypothesis. Consequently, by combining them in a suitable way better performance should be attained. A static case of track-to-track association and a dynamic case of measurement-to-track association are considered to evaluate the performance of the algorithm and the benefits are quantified.