On account of advantages like wide radar coverage and goodsheltered performance, passive tracking is playing an irreplaceable role inmodernized war. However, suboptimal performance or divergence of theresult may occur when using a nonlinear filter in a bias tracking systemwith unknown initial state and covariance of noise. Little literature isdealing with this problem. It is thus meaningful to come up with a jointregistration, association, and parameter estimation algorithm in field ofpassive tracking and design a data fusion and parameter estimationsimulation system with high expansibility and applicability.After referring to many advanced registration and associationalgorithms especially Expectation-Maximization algorithm which has arelatively better performance in the field of parameter estimation formixture models with latent variables, I proposed a novel joint registration,association, and parameter estimation algorithm for passive biasedtracking system. The ML estimation of complete data log is done byEM-UKS iteratively till parameter convergence. The association andregistration parameters are estimated in M-step, while the target states areupdated in the E-step by UKS. Simulation shows better estimationperformance than augmented UKF, and similar association performancewith three data fusion algorithm. Based on the analysis mentioned above, a multi-sensor multi-targetsimulation software system is built and described at the end of this thesis.Applicability and expansibility are guaranteed by its friendly UI design,server-client network structure, and open algorithm interface. |