In this paper we use a continuous wave radar to detect unmanned aerial vehicles (UAVs). The proposed method can be divided into two stages: suspected targets extraction and trajectories obtaining. In the first stage, we use a Double-GMM-Iteration method, in which two Gaussian mixture model background subtractors are applied to remove the background and obtain suspected UAV targets. In the second stage, we combine multi-frame information to obtain the suspected target trajectories through an improved track-before-detect (TBD) method. Finally, trajectories longer than the threshold are determined as U A Vs, and their relevant parameters are calculated. Experiments show that our method can effectively label the trajectories of UAVs, and the probability of false alarm is also low.