The Study Of Target Tracking Based On Feedback Neural Networks

Multi-target tracking and maneuvering target tracking are two important aspects of target tracking. Data association and tracking filtering are two chief issues in target tracking. Because the neural network has the ability of parallel processing massively, good adaptability, and powerful ability of learning and association, people introduce it to target tracking and achieve a lot. This proves the validity and importance of using neural networks to solve and optimize target tracking. And the feedback neural network has better performance in target tracking because of its better structure and dynamic characteristics.In multi-target tracking, data association is an extremely important problem, which is also a difficult problem. When the condition of multiple targets crossing in dense echoes environment emerges, Nearest Neighbor Algorithm and Probabilistic Data Associate Filter are in vain, more advanced ways are needed. This paper adopts Joint Probabilistic Data Associate Filter. With the targets and measurements increasing, the number of data association feasible events in JPDAF grows exponentially, appearing “combination” explosion. For that, this paper researches to use the noise-tuning-based hysteretic noisy chaotic neural network to solve the joint association probability. In MATLAB simulation, the effectiveness of this method was verified comparing with the noise chaotic neural network.In maneuvering target tracking, filter is a key. The motion of a maneuvering target is nonlinear, so it can not be described by one model. IMM Kalman filter can describe target motion better by using different models in different motion segments. But IMM Kalman filter also has its inherent shortcomings, such as insufficient targets and filtering error. In order to improve the tracking effect of maneuvering target tracking further, the paper uses Elman neural network to correct IMM Kalman filter. The simulation experiment was made using MATLAB. We can see from the tracking diagram and the root mean square error that the tracking effect is improved further. By being compared with the correcting effect of BP neural network and RBF neural network, the advantage of the feedback network like Elman neural network is reflected.