Joint tracking and classification of multiple extended targets via the PHD filter and star-convex RHM

Abstract

For joint tracking and classification (JTC) of multiple extended targets in the presence of clutter and detection uncertainty, this paper proposes a recursive algorithm based on the probability hypothesis density (PHD) filter and star-convex random hypersurface model (RHM), resulting in the JTC-RHM-PHD filter. By modeling the extent state via the star-convex RHM, the JTC-RHM-PHD filter can classify the extended targets using the feature information of size and shape, instead of the size information only in the random matrix model (RMM)-based JTC method. To integrate the prior class-dependent information into the tracking procedure and obtain the recursive expressions of the JTC-RHM-PHD filter, the relationship between the instantaneous extent state and prior class-dependent information is first constructed. Then, the target state is remodeled by two vectors: one for the kinematic state and another for the extent state. The kinematic state is estimated via a Kalman-like filter. For the extent state, the measurement-based extent state and class-based extent state are obtained via the sensor measurements and prior class-dependent information, respectively. The two kinds of extent states are used to determine the target class. At last, a Gamma-Gaussian-Gaussian mixture implementation of the JTC-RHM-PHD filter is presented. Simulation results indicate that: (1) Compared with the RMM-based JTC method, the proposed JTC-RHM-PHD filter can correctly classify multiple extended targets with similar sizes. (2) Compared with the RHM-PHD filter, the proposed JTC-RHM-PHD filter has superior tracking performance with a similar algorithm complexity.