Augmentation of the IMKF with KF-Trained Neural Networks for Adaptive Multiple Target Tracking

The Air Force Flight Test Center (AFFTC) has a compelling need for innovative research to develop on-line intelligent learning algorithms for Kalman filtering for tracking multiple highly dynamic targets. Even for the case of single-target tracking, challenges include the reduction of bias-errors, filter divergence, and the selection of error-states. Turbulent environments and other physical factors can contribute to nonlinear target motion thereby confounding standard Kalman Filter approaches. The proposed effort includes the investigation of means for extension of the Identity Management Kalman Filter (IMKF) by extending the model to associating tracking models to targets for robust target tracking of multiple highly dynamic targets. This includes extending the IMKF to enhance its performance for nonlinear tracking. The candidate systems shall employ neural networks for adaptive target-model enhancement and adaptive selection of the target model (for each target) that is fast enough for online near-real-time performance. The effort shall seek solutions that are applicable to both online tracking and offline trajectory reconstruction and also improve performance in cluttered sensor data so that they are robust enough to automate track reconstruction normally performed manually by engineering staff.