A Study Of Data Association, Multi-Dimension Assignment Problem In Multitarget Tracking

With the development of science and technology, especially the rapid development of sensor technology, computer technology and information technology, modern tracking environment is becoming more and more complex and modern tracking theory is also being developed and improved. Data association is a key technology of target tracking, especially for multitarget tracking, and it is also a hot subject all along. At the same time, the advantages of an imaging sensor make image-based tracking develop rapidly. And image-based tracking is a new subject of modern tracking. This dissertation studies some key problems in modern tracking technology. The main contributions are as follows:(1) In joint probability data association(JPDA) algorithm, each return can belong to only one target, and each target can own only one return. However, in dense multi-return environments and image-based tracking systems, a return may come from multiple targets, and a target may produce multiple returns. So the above assumption is not conformable to such a case. Consequently, we propose a new feasible rule, where the con-elation of returns and targets is considered to be a multiple-to-multiple problem. On this basis, we put forward an idea that a generalized joint event consists of two generalized events. And A Generalized Probability Data Association (GPDA) algorithm is given by using Bayes' rule. Additionally, the theory analyses show that the precision of the new algorithm is superior to that of JPDA, and the algorithm has much smaller computation burden than JPDA.(2) Taking the multitarget tracking for example, we analyze the performance of GPDA algorithm in various given tracking environments by using Monte Carlo simulation. For a measurement to multiple targets, we consider the case of imperfect measurement because of sensors' resolution in the environment of dense target formation and target across. For a target to multiple measurements, we consider the case of small target tracking by using an imaging sensor. Additionally, the dissertation studies the computation burden and computing memory, and simulation results demonstrate thecorrectness of theory analyses.(3) In this dissertation, we give a uniform model of multi-dimension problem by analyzing the standard model of assignment problem in operational research and the multi-dimension assignment model in data association. On this basis, we give three "pruning" theorems by transformation of the solution matrix using corresponding cost matrix, and give their theoretical proving process.(4) On the above basis, we propose a "priming" algorithm of multi-dimension assignment problem and analyze its mechanism and computation burden. The theoretical analyses and simulation show that the algorithm is easy to be implemented in computer and has smaller computation burden.(5) Using "pruning" algorithm, we give the assignment process of returns and targets in an example of passive multisensor multitarget tracking and give two examples of cost to resource assignment problem. The analysis results show that the "pruning" algorithm has good real time ability for multi-dimension assignment problem.(6) Combining interacting multi-model(IMM) with fuzzy logic, we propose an image-based fuzzy IMM tracking algorithm and give the whole frame of the algorithm. In the algorithm, we combine the measurements from a conventional sensor with that of an imaging sensor to constitute a mix measurement. And the Markovian switching, which is used in IMM, is replaced by fuzzy rules. Moreover, the data association is implemented by using fuzzy logic.