Neural Network Aided Unscented Kalman Filter for Maneuvering Target Tracking in Distributed Acoustic Sensor Networks

Abstract

A new neural network aided unscented Kalman filter is presented for tracking maneuvering target in distributed acoustic sensor networks. In practice, the system dynamics of these problems are usually incompletely observed, there may be large modeling errors when the target is maneuverable and some parameters of the system models may be inaccurate. So we propose using an offline trained neural network to correct these errors, the nonlinear inferring process is done by the normal unscented Kalman filter. This method doesn't need complex modeling for tracking maneuvering target and is very suitable for real-time implementation because the implementation time is only the sum of the unscented Kalman filter and the neural network recall time