Multisensor Track-to-track Association And Bias Removal In Complex Environments

Mutisensor-multitarget tracking, a typical application of data fusion, plays a fun-damental role in the network-centric operations of the information era. It attracts a greatdeal of attention from the academia and industry. Most of the existing data fusion systemsfocus on suppressing the impact of random noises, but in engineering practice, they couldnot work well in the presence of coexisting random noises, sensor biases, false alarms andmissed detections. In real-life applications, the complex interference factors, especiallythe sensor biases, make the position state of targets unreliable, and tend to cause misas-sociations. While, misassociations will destroy the process of estimating and correctingsensor biases. In view of the problem that the existing fusion system can not work stablyin the complicated practical environment, the thesis focuses on the coupled sensor biasestimation module and track-to-track association module, and launches the research fromthe following three aspects:First, analytic performance prediction of track-to-track association in complex envi-ronments. We derive the analytic expressions of the correct association probability for theglobal nearest neighbor association algorithm. The intrinsic impact of the key scenarioparameters (mainly including sensor biases, random noises, false alarm spatial densityand target spatial density) on the association performance is uncovered. The simulationresults match the analytic prediction results quite well.Second, this thesis puts forward a series of robust sensor bias estimators to solve theproblem that the performance of traditional sensor bias estimation approaches degradesseriously in complex environments. Specifically, to conquer the impact of misassocia-tions, a robust sensor bias estimator based on the least median of squares is proposed.Then, based on the special properties of misassociations, we propose the least quantileof squares estimator, which promotes the ability of tolerating misassociations. To over-come the impact of ill conditioning caused by the dense-target and (or) dense-sensorscenarios, a robust sensor bias estimator based on the bounded-variables least squares isproposed, which enhances the numerical stability of the estimation results. Finally, thebounded-variables least median of squares estimator is proposed to restrain the impact ofmisassociations and ill conditioning simultaneously.Third, to solve the track-to-track association problem in the presence of sensor bi- ases, the thesis does the research in two directions. From the perspective of constrainingthe efect of misassociations, the single-start robust iteration algorithm is proposed. Thisproposed algorithm efectively inhibits the negative impact of misassociations, by makinguse of the robust bias estimator in the process of the alternate iteration of sensor bias es-timation and track-to-track association. Then, in view of evaluating the association resultwithout ground truth, the concept of consistent association number is presented and thealgorithm based on maximizing the consistent association number algorithm is proposed,which reduce the sensitivity of the robust alternate iteration process to the precision ofsensor bias estimation. From the perspective of removing the efect of sensor biases, therobust track-to-track association algorithm based upon the reference topology (RET) fea-ture is refined. The rigorous mathematical definition of RET is presented according tothe set theory. The robustness of RET to sensor biases is analyzed theoretically. The setdistance metric between RETs is modified. The concept of topology number is proposedto control the computation complexity of the algorithm. Simulation results verify theefectiveness of the proposed algorithms.