Reconstruction of a target's location in time and space is crucial to the functionality of a tracking system used in warfare. One commonly-used approach to position and velocity estimation is Kalman filtering; however, the traditional Kalman filter has problems relating to bias errors, filter divergence, and filter parameter tuning. Artificial Neural Networks (ANNs) can be used to solve these problems through covariance estimation of input noise or by making predictions from a training set. Phase I of this research and development effort is to analyze the feasibility of an ANN-based online adaptive Kalman filter and its alternatives; a variety of solutions will be developed in order to optimize the target-tracking problem for different criteria. In Phase II, a working prototype using one or more of these solutions will be implemented to track multiple targets in near-real-time. Systran Federal Corp. (SFC) has assembled a distinguished team to address this proposal. In conjunction with our research partner, University of Missouri-Rolla (UMR), we are proposing a novel and innovative approach to meeting the demanding requirements listed in the solicitation. At the conclusion of Phase II, we plan to have a pre-production version of our product ready for immediate deployment in selected applications.
Keywords: Tracking, Intelligence, Artificial Neural Networks, Algorithm, Kalman Filtering, Machine Learning, Adaptive Filters