Research On Multi-target Tracking Information Fusion Based On Neural Networks

Data fusion is a rising multidisciplinary technology. Its application technology deals with a wide range of disciplines and technologies, so it is one of the most potential and valuable research directions in the world today, for it can strengthen system viability, reduce the information fuzziness, promote system reliability and so on. Target Tracking, which is a major part of data fusion, includes single target tracking and multi-target tracking. In practice, it tends to have many indefinite problems, so that data are so complex that data association and state estimation become very difficult. Therefore, it's inevitable to introduce Neural Networks and fuzziness system into Multi-target Tracking technology to meet people's increasing requirement on behavior of data fusion system since it's hard to work just depending on only one fusion method.In this dissertation, multi-target tracking information fusion algorithm based on Neural Networks has been studied for the purpose of promoting system reliability and improving its robustness. First of all, the basic theorems, topology and algorithms in common use of multisensor information fusion have been studied, and target tracking theory and its process have been detailed introduced. Furthermore, Neural networks-based tracking theory is combined with fuzziness theory in this paper, and neural networks applications on multi-target have been researched deeply. Since the study rate of Elman neural network has a great effect upon convergence speed and stability, an improved adaptive study algorithm of Elman network is put forward, which can change study rate according to twice sequent variation of training error. Based on researching the structure, principle and study method of Elman neural networks and neural networks-based tracking, a new multi-target tracking method based on Elman neural network is presented. At last, the method mentioned above is applied to adopt two sensors tracking two targets, then Matlab7.0 has been used to get computer simulation. Simulation result shows that the method based on Elman network can still track efficiently when two targets positions are very close. Under the strong maneuver condition, the distance needed to track the target again applied the new method is shorter than the one based on BP neural network, the final result is more ideal.