An initialization clustering method for SMC-PHD based on likelihood function

The sequential Monte Carlo probability hypothesis density (SMC-PHD) filter has been demonstrated an elegant multi-target tracking algorithm which is suitable for highly nonlinear systems. However, as a typical technique of extracting multi-target state for SMC-PHD, standard K-means clustering is unreliable due to the initialization, which chooses the initial cluster centers at random. To solve the problem of the K-means clustering, an initialization method for SMC-PHD based on likelihood function is proposed in this paper. First the likelihood function based statistic is defined and the selection of threshold used to obtain the target-generated measurements is illustrated. Then a novel initialization method is discussed in detail. Finally simulations are presented. The results show that the proposed method has better performance than the standard K-means clustering.