Target State Estimation Based on the Combination of Deep Learning and Unscented Kalman Filter

This article studies a target state estimation method based on Unmanned Aerial Vehicle(UAV) observation images, focusing on the problem of unknown target motion models and observation noise interference. We propose a DeepUKF algorithm that combines Long Short Term Memory (LSTM) and Unscented Kalman Filter (UKF). The algorithm uses the LSTM network to learn the target state transition model and system observation noise and then uses them as inputs for the UKF algorithm. This method effectively combines the feature learning ability of deep learning with the interpretability and robustness of the Kalman filter. Through simulation experiments on tracking maneuvering targets, the DeepUKF algorithm has shown excellent performance in the accuracy and stability of target state estimation.