This paper proposes an efficient approach to detect dynamic objects in a range image projected from LiDAR point cloud data. Between an original frame and its transformed frame, neighboring features are used to efficiently generate the initial flow seeds first. Motion consistency is checked to refine the flow seeds in the background. Scene flows are then used to verify the forward-backward consistency of candidate dynamic pixels between adjacent frames, accurately removing false dynamic regions. Finally, dynamic objects are post-processed using clustering, region growing, and object consistency checking. Experiments and evaluations demonstrate that our method surpasses other approaches that rely solely on 3D point clouds for dynamic object detection.