The partitioning method based on distance does not work well when the densities of measurements from different extended targets are very different, and thus reduce the performance of the extended target probability hypothesis density (PHD) filter. Based on the share nearest neighbors(SNN) similarity, this paper presents a measurement set partitioning approach, which could work well in the situation that the densities of measurements from different extended targets are very different and further could enhance the tracking performance of the filter based on the proposed method.The SNN similarity that incorporates the neighboring measurement information is introduced instead of the distance between measurements used in the measurement set partitioning, and thus it is relatively insensitive to variation in measurement density. Although calculating the SNN similarity consume some time, the resulting PHD filter based on the proposed partitioning approach does not cause more computational burden due to the lesser number of the resulting partitions. Especially in high clutter scenarios, a significant reduction in computational complexity can be achieved. Simulation results demonstrate the superiority of the filter based on the proposed partitioning approach.