The Research Of Data Fusion For Multi-target Tracking Based On Neural Network

Content: Data fusion is a comprehensive subject of studying the processing of multi-source information. Target tracking is an important part of data fusion. However, there is a contradiction between increasing rapid response and improving the tracking precision in traditional multi-target tracking technology. In order to get better solution, many experts have been studying in this area.Since the 1980s, with the revival of neural networks, the multi-target tracking technology has greatly developed. At the beginning of this paper, properties, the structure of the neural network are analyzed, and an algorithm for multi-target tracking based on kohonen neural network is proposed, which integrates the neural network and the multi-target tracking technology. This algorithm uses position information as the input of network, and it is applied to multi-target tracking by means of the kalman filter algorithm and other methods. Simulation results show that the algorithm can realize precise tracking in certain condition. However, because of the limitation of the kohonen network itself, it may not be able to meet the real-time requirement. To solve the problem, the fuzzy kohonen clustering network (FKCN ) is introduced here and compared with kohonen network, accordingly, an improvement tracking algorithm is proposed. Simulation results show that the improvement algorithm based on FKCN can enhance the network convergence rate and satisfy the request of real-time. Combining FKCN algorithm with the idea of" filter after fusion", a multi-sensor fusion algorithm is proposed. Simulation results also show that this algorithm works well in multiple targets condition and it provides a new way of the data fusion for multi-target tracking field. Finally, in view of existing problem in simulation experiments, the self-adaptation tracking algorithm based on BP neural network is studied, and its improvement schemes are proposed.