Research On Target Robust Modeling And Tracking Algorithm

Target Tracking is a classical problem in the area of dynamic filtering, its target is to track the interesting target using certain measurements and tacking algorithm. How to model the target using a precise model and how to cope with the outlier in the measurements are two difficult problems in tracking system, which is the robustness of the system in essence. Our thesis has researched the robustness of the tracking system in the following two ways:1) Explorer how to establish an accurate dynamic model for targets.2) To cope with the outlier in the system in order to improve the robustness of the tacking system.Firstly, in the single-target tracking system, to cope with the inaccurate target dynamic model, a new method of modeling the dynamic model---robust dynamic modeling method based on polynomial prediction model is proposed, and we also give the interactive dynamic filtering algorithm. The proposed modeling method uses the random walk model, first-order and second-order polynomial prediction model as the model set, its purpose is to use the idea of interacting multiple model, and use Markov probability transition matrix to give each model a probability and then mixed the models in the model set to obtain the true target dynamic global model, thus to improve the adaptive dynamic target capacity and enhance the robustness of the algorithm. The computer simulations verify the effectiveness and practicality of the algorithm for maneuvering target tracking.Secondly, we extend the modeling idea above to the estimation of the instantaneous frequency of the chirp signal modeling, and then use the UKF (Unscented Kalman Filter, UKF) to estimate instantaneous frequency of the chirp signal. This has opened up a new field of application to achieve a real-time tracking of the instantaneous frequency. Analysis and simulation results show that, the proposed algorithm has better robustness than the existing single-model tracking algorithm when there are instantaneous frequency transitions.Finally, in multi-target tracking, when the target dynamic model is accurate, but there are outliers in the measurement, a robust variable-lag PHD (Variable-lag Probability Hypothesis Density, VPHD) smoothing algorithm is proposed. This method uses the forward PHD filtering to obtain a priori information to adjust the smoothing interval, which can overcome the problem of the fixed-lag PHD (Fixed-lag PHD, FPHD) smoothing algorithm. Computer simulations in multi-target scenarios show that our proposed algorithm is more robustness, and can both improve the accuracy of state estimates and target number estimates.