| Multi-target tracking theory has been widely used in the filed of military and civilian, but it is also the important and difficult problem which is focused by many subjects and fields. In dealing with this problem, the traditional methods use the data association step at first, and then evaluate the state of the targets or the parameters. Data association is the key step in the tracking problem. When the number of the targets is too many and there are lots of clutters and false alarm, data association is very difficult to process and a lot of problems are appeared such as combination explosion and calculation increasing as exponential type. Random sets is a method which is dealing with the target tracking problem avoiding data association. But there is integral operation of set function in the calculation procedure, and there is no analytical solution in common sense. Therefore, a problem is generated that is hard to implement in the reality because of the complex calculations. Along with the deep going research, some scholars provide the PHD filter based on the frame of random sets theory. This method simplifies the sets function integral operation to the single variable integral operation, which has the possibility to realize the numerical calculation. Although PHD method has some advantages, there exist some problems in its application. For example, PHD can only evaluate state of the target, but it can not get the trajectory of the targets. The numerical method is needed to compute function integral operation in the PHD for the most part, but how to get a more efficient numerical method is a problem deserved to research. PHD is not appropriate for all the tracking problems, therefore, how to compare PHD with other classic tracking methods and decide their suitable ranges is a problem eager to be solved.The research of the paper is aiming at the problems exist in the theory and application of PHD, and the main contents are listed as follows.1) After reviewing some domestic and foreign literatures and analyzing the advantages and disadvantages of various classical algorithms, PHD method is provided as a research direction in the future which can overcome shortcomings of some other methods and solve the target tracking problem better.2) GM-PHD filter and its application in multi-target tracking. In the gaussian mixture situation, an improved GM-PHD analytic calculation algorithm is proposed. In this algorithm, a'nearest neighbor'clustering algorithm is used to get the trajectory of the targets after getting the estimated state of the target sets.3) The performance analysis and comparison between GM-PHD and JPDA. At first, general analytical forms to evaluate calculation complexity of each algorithm are formulated by analyzing and totaling their major operation steps. And then the calculation complexity of two algorithms is compared through three cases respectively, which are divided on the basis of associated complexity between states and the measurements. The suitable range of each algorithm is shown at last.4) UPF-PHD filter and its application in the multi-target tracking. From the view of numerical calculation, a new PHD algorithm is proposed, which combines UPF to undertake numerical calculation, and the'nearest neighbor'clustering thought is used to solve the trajectory of the targets. |
PDF Full Text Request