Track splitting technique for the contact lens problem

The contact lens problem in tracking has severe measurement nonlinearity that will cause consistency problems and large errors for existing nonlinear filtering techniques including the EKF, the UKF and the particle filter. In our previous work, the proposed measurement covariance adaptive (MCA) extended Kalman filter (MCAEKF) was shown to be consistent and have superior tracking accuracy. The only drawback of the filter is that it has loss in accuracy in the early stages of the filtering due to the artificially enlarged measurement noise covariance. In this paper, a novel track splitting technique is proposed to divide an inaccurate track into a set of sub-tracks that are accurate enough such that the linearized EKF consistency requirement is satisfied for each sub-track. Simulation results show that the proposed track splitting EKF (TS-EKF) approach can effectively prevent filter divergence and has no loss in range accuracy in the early stages of filtering.