Radar tracking of a maneuvering ground vehicle using an airborne sensor

In this paper we compare four different sequential estimation algorithms for tracking a single maneuvering target using data collected by an airborne sensor. The target is ground-based and its motion can be modeled according to Markov chains theory. Maneuvers refer to an inertial reference system and are defined by three different kinematic models: stop, constant speed and maneuver. We analyze a realistic car traffic scenario by considering a sensor whose motion is circular around the designated target. The target motion is defined in Cartesian coordinates, while measurements are expressed in sensor-centered spherical coordinates. Both the target and measurement update equations are characterized by the presence of additive Gaussian noise with known powers. The particular geometry between the target and the sensor can introduce fictitious accelerations. As a consequence, heavy nonlinearities can be generated, especially during the stop and turning phases. This problem is addressed defining both the target and sensor motion directly in continuous-time. In order to extract the kinematic features of the target, Bayesian inference is made on the set of noisy measurements. A special interest is devoted to the use of a particle filter (PF). In particular, we compare two PF-based algorithms, i.e. the multiple model particle filter (MM-PF) and the multiple model auxiliary particle filter (MM-APF), to the well-established extended Kalman filter (EKF) and the interacting multiple model EKF (IMM-EKF). Advantages and disadvantages of the proposed algorithms are illustrated and discussed through computer simulations