4.4.1. Overview

There are several different types and solutions for tracking objects [18]. Kalman-filter based methods are widely used for target position tracking.

For tracking objects with high manoeuvring capabilities, utilisation of the Interacting Multiple Model (IMM) filter is a good practice. Although the IMM filter is a well known approach for object tracking, let us briefly point out the basic principle. The IMM filter considers multiple motion models (e.g., constant velocity, constant acceleration, constant turn rate models) each associated with a dedicated Kalman filter. The Kalman filters are running simultaneously in parallel and their outputs are blended to generate the estimated state of the system according to the likelihoods of being in a certain motion mode. The higher the probability of a mode, the higher its contribution to the blended state. The state of a more probable mode is affected slightly by less probable modes [19,20]. During this process, the likelihoods of being in a certain mode (e.g., constant velocity mode) and the likelihoods of transitions between modes are calculated based on the last state. In order to reduce the transient period every filter is reinitialized with the mixed estimate of state and covariance [21].
