Joint Multiple Target Tracking and Classification Using Controlled Based Cheap JPDA-Multiple Model Particle Filter in Cluttered Environment

In this paper, we address the problem of jointly tracking and classifying several targets in cluttered environment. It is assumed that the motion model of a given target belongs to one of several classes. We propose to use the multiple model particle filter (MMPF) to perform nonlinear filtering with switching dynamic models. Moreover, the principle of joint probabilistic data association (JPDA) is used to determine the measurements origin. Besides, the joint probabilities are calculated using Fitzgerald's had hoc formulation (Cheap JPDA) whose efficiency has been proven in the literature. On the other hand, a controller based on the quality of the innovation has been implemented in order to tune the number of particles. The feasibility and the performances of the proposal have been demonstrated using a set of Monte Carlo simulations dealing with two maneuvering targets.