Research On State Estimation And Data Association Of Motion Targets

It is of vital importance to improve the precision of tracking for maneuvering targets. Typical issues in tracking for maneuvering targets include state estimation and data association. The precision of tracking for maneuvering targets depends largely on how much better maneuvering model, suitable model-set, robust filtering algorithm, and exact data association.This dissertation deals with above four challenging topics. At first, by analyzing the characteristic of traditional "current" statistical model and its covariance adaptive algorithm, we hunt how to improve its performance by adaptively adjust the parameters of model and overcome inner shortcoming in the framework of Kalman Filter. What's more, we study how to track turn targets with Curvilinear Model and Constant Turn. Secondly, three classes of general design methods for model set are introduced, then number-theretic method is described in detail and used to design model set used in the application of the multiple-model approach to estimate the states of targets. Thirdly, after introduced three classes of primary filters for nonlinear problems, the sampling methods of Unscented Kalman Filter and its compound filtering approach are studied. At last, the detection methods of multiple-targets in monopulse radar and data association methods are elaborated, then how to improve the precision of multiple targets tracking with the results of targets detection is presented.The main research and contributions of this dissertation include:1. An adaptive strong tracking approach based on modified current statistical model is proposed, by adaptively adjusting the parameters of model and overcoming the shortcoming of Kalman Filter, in order to enhance the performance of system for tracking sudden maneuver while maintains the better performance of tracking commonly motion. At first, in algorithm, the motion state of target is divided into two different motion degrees by setting the detection thresholds of motion according to probability distribution of statistical distance of measurements innovation so as to adjust the parameters of model and waiting factor of filter accordingly. The match degree of motion model and system mode is enhanced by adjusting the former, but the shortcoming of tracking sudden maneuver existed in Kalman Filter is overcomed in a degree by adjusting the latter.2. We proposed two adaptively tracking approaches for turn targets, which include an adaptive approach based on Curvilinear Model and a two-layer IMM approach based on Constant Turn Model. With the physical relations between the heading angle of velocity, angular velocity and angular acceleration of motion target, the angular velocity is obtained while the angle of velocity is filtered with multiple models. At last, the filtering method for heading angle of velocity is combined with Curvilinear Model or Constant Turn Model, in a result, the tracking performance of turn target is enhanced and the angular velocity is estimated exactly.3. With the guidance of minimizing probability distribution mismatch, a new design method for model set using represent points of probability distribution with number-theoretic methods is proposed. Model set used in a multiple-model approach is composed of representative points which include F-discrepancy (or quasi-F-discrepancy) and mean square error representative point represented probability distribution function of system mode can be obtained with number-theoretic method.4. We generalize the minimizing symmetric sampling approach using 2n+l samples to a new symmetric ones using 2kn+1 samples to improve the filtering performance with Unscented Kalman Filter by using more samples. What's more, we presented two major directions to improve filtering performance with Unscented Kalman Filter for nonlinear problem: one is a traditional method that Unscented Kalman Filter is combined with alternative filter or innovation filter, the other is a compound filtering way that Unscented Kalman Filter is combined with other nonlinear filter, which may include sequential compound filtering approach and interactive compound filtering approach.5. We present that joint probabilistic data approach combined with the detection technique of multiple targets in monopulse radar in the process of multiple targets tracking. In a result, the applied scope of association approach can be expanded and the decision of target tracking can be improved since the probabilistic limitation of feasible event is overcomed in association approach by determining the origination of measurements.