<p>Dynamic point clouds can be compressed by eliminating spatial and temporal redundancy, but few research studies have considered both simultaneously. Existing research can only distinguish the specific foreground and background of the scene, achieving the former with a limited compression ratio, while the latter relies on 3D motion estimation with high computing complexity. Therefore, it is difficult to meet the real-time requirement of building remote human-robot interaction system. Aiming at the shortcomings of the above methods, a dynamic point cloud fast compression framework based on eliminating spatial and temporal redundancy is proposed. On the one hand, for eliminating the spatial redundancy, the information of the recognized objects in point cloud data will be replaced with their categories and poses. On the other hand, we eliminate the temporal redundancy by using the motion vector of the point cloud macroblock. Specifically, in order to reduce cost time of the 3D motion estimation, we firstly introduce the 2D image ORB feature to calculate the motion vector from current frame to key frame, and then look for the similar macroblock in key frame. Thus, the information of the macroblocks similar to the key frame in the current frame is replaced by the index of the corresponding macroblocks of the key frame and the motion vector. Experimental results show that the proposed method has higher computational efficiency and compression quality than the existing compression algorithm.</p>