Object Identification and Tracking via Noise Updated Iterative Extended Kalman Filter

The paper considers the problem of simultaneous identification and trajectory tracking of moving objects (either 2D or 3D) from moving sensors. The identification is parametric and is based on knowing the family that the object belongs to, e.g. ball, ellipsoid, box, etc. The mathematical formulation results in implicit measurements, i.e. an algebraic equation that includes both state variables and actual measurements. The method of solution is via Extended Kalman Filter where the unknown parameters are regarded as additional state variables. Standard Extended Kalman Filter and Iterative Extended Kalman Filter yielded unsatisfactory results, mainly due to the nonlinearity of the measurements in both the state vector and the noise. A new algorithm, called Noise Updated Iterative Extended Kalman Filter is suggested. Its deviation from the standard iterative Kalman filter is in estimating the measurement noise at each iteration. The estimated noise is then used in the linearization stage to obtain a more accurate linear approximation. The method has been applied to the online identification and tracking problem, with substantial improvement in performance.