The typical sequential Monte Carlo(SMC) implementation of probability hypothesis density(PHD) filter is similar with the particle filter. Both of them make use of a large number of particles to estimate the multiple target states. The fixed number of particles is assigned for each expected target in typical SMC implementation, which will result in larger time cost of the algorithm. Therefore, the SMC implementation based on Kullerback-Leibler divergence(KLD) is proposed. The KLD is computed for the two particle sets in different sizes. Then, the KLD is compared with the pre-threshold to obtain the number. The number of particles can be adaptively adjusted in the proposed implementation. Simulation results show that, the proposed implementation can improve the tracking efficiency, which is superior to the typical implementation in most time steps.