In this paper, a novel target recognition based clustering algorithm is proposed for time-varying channels. Power angle spectrum (PAS) is extracted from measurement data by using Bartlett beamformer. Then the clusters in the PAS are separated from the background by using the proposed algorithm, where the amplitude distribution of the elements in the PAS is considered. Moreover, morphology operations are applied to further divide the clusters which are connected to each other. It is found that, the dominating clusters in both line-of-sight (LoS) and non-line-of-sight (NLoS) environments can be well recognized by the proposed algorithm with low computation cost. By using the proposed algorithm, the dynamic changes of the clusters in real-time channel measurement can be clearly observed, without using any high-resolution parameter estimation.