Research On Methods For Multiple Extended Targets Tracking With Unknown Measurement Noise Based On Random Finite Set

For the extended target, since each target will produce several measurements in every sampling period, if we want to associate target with its corresponding measurements, it will be a big problem. To study a kind of real and effective tracking method has important practical significance and value. In recent years, multitarget tracking method based on random finite set(RFS) has been accepted because of its low computation and high ability of solving problems. This paper mainly focus on the extended target tracking method based on RFS. The main contents of the work are summarized as follows:1. Extended target tracking method based on Gaussian inverse Wishart PHD(GIW-PHD) filter is introduced, the filter presented in this paper is capable of estimating both the kinematic state and extension state with known measurements noise covariance. The target kinematical state are modeled using a Gaussian distribution while the target extension is modeled using an inverse Wishart distribution. When the measurements are accepted, they can be used to update the parameters in Gaussian distribution and inverse Wishart distribution, so that the information such as target location, velocity and size can be tracked.2. Extended target tracking method based on Random Hypersurface Model PHD(RHM-PHD) filter is introduced. Like GIW-PHD, the method is not only under the condition of known measurements noise covariance but also takes extension state into account, however, it has difference in measurement modeling with GIW-PHD. In RHM-PHD filter, the measurement is made up of the noise and measure source distributing on the target surface while in GIW-PHD filter, the measurement is made up of the noise and kinematic state of target. Moreover, parameters in extension state are embedded into the kinematical state, estimation result of target shape, size and orientation can be obtained by updating the kinematical state.3. Extended target tracking method based on variational Bayesian CBMe MBer(VB-CBMe MBer) filter is proposed. The method’s advantage is that it applies to the situation of unknown measurements noise covariance, meanwhile, it also comes up with a new kind of measurement modeling. The core idea of the method is to approximate the joint probability density of measurement producers’ state and unknown measurement noise covariance with variational bayesian technique, and then embedded into the framework of CBMe MBer, at last, using clustering algorithms to get the extended target state after tracking measurement producers.4. Extended target tracking method based on variational Bayesian PHD(VB-PHD) filter is proposed. This method still applies to the situation of unknown measurements noise covariance and the new measurement modeling method in VB-CBMe MBer, however, it approximates joint posterior intensity that differs from the latter. In VB-PHD filter, the measurement producers’ state can be obtained after updating the parameters in the approximate distribution, clustering algorithms then can be used to get the extended target state. The simulation experiment show that the method has a similar result with CBMe MBer with a true measurement covariance.