Abstract:
Target tracking algorithms adopted in modern radars are designed such that they can track multitarget by considering target births and target deaths. These algorithms are derived by integrating the data association techniques into the single target lters. Recently, target tracking methods exploiting random nite sets have been emerged as an alternative to the data association techniques. Unlike data association methods, random nite set based techniques do not perform tracking based on the targets but instead propagate a target intensity function covering the entire state space in time and thereby decrease the dimension of the state space. In this thesis, rstly on a linear scenario we investigate the e ects of receiver characteristics and variation of target intensity on the performance of PHD lter that is a random nite set based lter. The parameters we consider for receiver characteristics are detection probability and false alarm intensity; for variation of target intensity we investigate the e ect of target birth and death probabilities. We also provide a linear regression model representing e ects of these parameters. As a tracking performance metric we use OSPA distance. At each step, we compare our results with a data association method, global nearest neighbor technique, in order to identify the advantages and disadvantages of the both of the methods. Secondly we investigate the e ect of the nonlinear models on both of the methods. By xing the parameters to the values that results in equal average OSPA distances of both techniques in the linear case, we include nonlinear models in order to identify which technique is e ected by nonlinearity more.