The stable detection and tracking of moving objects are crucial for the autonomous navigation of Unmanned Ground Vehicle(UGV). In order to obtain information related to object motion, some methods have started to use the latest frequency-modulated continuous-wave (FMCW) LiDAR for detecting and tracking moving objects. Compared to traditional time-of-flight (ToF) LiDAR, FMCW LiDAR provides additional Doppler velocity information combined with 3D point clouds, allowing researchers to focus on dynamic objects through static-dynamic separation. However, previous studies often reused TOF LiDAR methods, neglecting the unique characteristics of Doppler velocity and the differences after point cloud separation, resulting in limited accuracy improvements. Therefore, this paper proposed CentroidTrack, a centroid-based method for joint 3D object detection and tracking. By using the centroid as the detection key, CentroidTrack avoided the severe loss of center features after separating static and dynamic objects. The method integrates motion estimation into the network model’s predictions, leveraging motion information from the Doppler velocity dimension in point clouds for efficient and streamlined motion estimation. Extensive experiments on four datasets demonstrated that CentroidTrack significantly outperformed baseline methods.