A comparison of a possibilistic and a probabilistic classifier in a multitarget tracking environment

Two different approaches to target classification and the representation of the uncertainty in the classification are compared. First, a probabilistic (Bayes) classifier is described that minimizes the average cost of the classification. Then, a possibilistic classifier is presented that minimizes the maximum possible cost given the possibility distributions of the attributes. An evaluation of the performance of both classifiers shows the sensitivity to deviations from a priori knowledge that is employed. This sensitivity is particularly important in military scenarios where we deal with intelligent adversaries. Finally, the application of the probabilistic and possibilistic classifier to the plot-track association problem in a multitarget tracker is demonstrated with real radar data.