Abstract
To track multiple extended targets for the nonlinear system, this paper employs the idea of the particle filter to track kinematic states and shape formation of extended targets. First, the Bayesian framework is proposed for multiple extended targets to jointly estimate multiple extended target state and association hypothesis. Furthermore, a joint proposal distribution is defined for the multiple extended target state and association hypothesis. Then, the Bayesian framework of multiple extended target tracking is implemented by the particle filtering which could release the high computational burden caused by the increase in the number of extended targets and measurements. Simulation results show that the proposed multiple extended target particle filter has superior performance in shape estimation and improves the performance of the position estimation in the situation that there are spatially closed extended targets.
Highlights
In most multitarget tracking applications it is assumed that each target produces at most one measurement per time step
We compare the performance of the proposed multiple extended target particle filter with the ET-RHMGMPHD filter and the ET-random hypersurface model (RHM)-sequential Monte Carlo PHD (SMCPHD) filter
The ETRHM-GMPHD filter uses the framework of the GaussianMixture probability hypothesis density (GMPHD) filter to avoid the data association [17] and applies the random hypersurface model describing the measurement source distribution for convex-star extended target [5] and employs the particle filter to deal with the nonlinear measurement model in update step
Summary
In most multitarget tracking applications it is assumed that each target produces at most one measurement per time step. The Gaussian process measurement model is integrated in the recently developed labeled multi-Bernoulli filter for extended objects in [22] In theory this kind of methods needs to consider all possible partitions of the measurement set and is computation-intensive though some algorithms have been proposed to consider only a subset of all possible partitions [16, 17]. The other is those approaches which extend the data association algorithm applied in tracking point targets to track multiple extended targets.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
