Human Pose Estimation Using Millimeter-Wave Radar: Prototype, Dataset, and Method

Jiang Liu; Yan Huang; Yanjun Zhang; Chengchang Tian; Zhen Yang; Kaihang Zheng; Rui Zhou; Hui Zhang; Jixin Chen; Wei Hong

doi:10.1016/j.eng.2026.10.001

Engineering ›› :202610001 DOI: 10.1016/j.eng.2026.10.001

research-article

Human Pose Estimation Using Millimeter-Wave Radar: Prototype, Dataset, and Method

Author information +

History +

PDF (5893KB)

Abstract

Three-dimensional human pose estimation (3D-HPE) is important for human-centered sensing, but vision-based solutions are limited by privacy concerns, illumination changes, and occlusion. Although millimeter-wave (mmWave) radar provides a privacy-preserving alternative, existing radar point clouds are often sparse and incomplete, and existing radar 3D-HPE datasets remain limited in subject scale, frame volume, and motion diversity. This paper presents a mmWave radar prototype for dense point-cloud sensing, a large-scale mmWave radar dense human pose estimation (mmWave-DensePose) dataset, and the part-aware spatio-temporal network (PAST-Net) for radar-only 3D-HPE. Specifically, we design a four-chip cascaded mmWave radar system with an azimuth and elevation resolution of and respectively, to generate dense human-related point clouds. The constructed dataset contains 101 subjects, 19 designed motions, numerous free motions, and 570 124 valid synchronized frames. Each valid frame contains 164 valid subject-related radar points on average, exceeding the representative released datasets and providing clearer body-region cues for radar-based 3D-HPE. The proposed method uses only mmWave radar point clouds as input and Kinect V2 skeletons as ground-truth annotations. PAST-Net reorganizes frame-level point-cloud features into skeleton-aware part-token representations and models their temporal evolution through part-wise self-attention, compensating for weak and incomplete observations in local body regions. Experiments on the proposed benchmark show that PAST-Net achieves an average mean localization error of 4.08 cm, reducing the error by 22.14% compared with a state-of-the-art baseline. Ablation results further confirm the contributions of the proposed modules and input features, which include skeleton-aware spatial modeling, part-wise temporal modeling, and radar-specific point attributes. These results demonstrate that the designed mmWave radar prototype and the proposed method provide an effective sensing basis for privacy-preserving and fine-grained 3D-HPE.

Keywords

Millimeter-wave radar / Three-dimensional human pose estimation / Millimeter-wave radar dense human pose estimation dataset / Skeleton-aware representation / Spatio-temporal modeling / Dense point cloud

Cite this article

Download citation ▾

Jiang Liu, Yan Huang, Yanjun Zhang, Chengchang Tian, Zhen Yang, Kaihang Zheng, Rui Zhou, Hui Zhang, Jixin Chen, Wei Hong. Human Pose Estimation Using Millimeter-Wave Radar: Prototype, Dataset, and Method. Engineering 202610001 DOI:10.1016/j.eng.2026.10.001

登录浏览全文

4963

注册一个新账户忘记密码

References

Publishing order | Descend order by publishing year | Descend order by cited within

[1]	Neupane RB, Li K, Boka TF . A survey on deep 3D human pose estimation. Artif Intell Rev 2024; 58: 24.

[2]	Wang J, Tan S, Zhen X, Xu S, Zheng F, He Z, et al. Deep 3D human pose estimation: a review. Comput Vis Image Underst 2021; 210: 103225.

[3]	Zheng C, Wu W, Chen C, Yang T, Zhu S, Shen J, et al. Deep learning-based human pose estimation: a survey. ACM Comput Surv 2023; 56(1): 11.

[4]	Zhou X, Li S, Liu J, Wu Z, Chen YF . Construction activity analysis of workers based on human posture estimation information. Engineering 2024; 33: 225-36.

[5]	Wang C, Xie S, Li K, Wang C, Liu X, Zhao L, et al. Multi-view point-based registration for native knee kinematics measurement with feature transfer learning. Engineering 2021; 7(6): 881-8.

[6]	Liu H, Zhang Z, Jiao Z, Zhang Z, Li M, Jiang C, et al. A reconfigurable data glove for reconstructing physical and virtual grasps. Engineering 2024; 32: 202-16.

[7]	Andriluka M, Pishchulin L, Gehler P, Schiele B 2D human pose estimation: new benchmark and state of the art analysis. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2014 Jun 24-27; Columbus, OH, USA. Piscataway: IEEE; 2014. p. 3686-93.

[8]	Ionescu C, Papava D, Olaru V, Sminchisescu C . Human3.6M: large scale datasets and predictive methods for 3D human sensing in natural environments. IEEE Trans Pattern Anal Mach Intell 2014; 36(7): 1325-39.

[9]	Cao Z, Simon T, Wei SE, Sheikh Y . Realtime multi-person 2D pose estimation using part affinity fields. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2017 Jul 21-26; Honolulu, HI, USA. Piscataway: IEEE; 2017. p. 7291-9.

[10]	Zheng C, Zhu S, Mendieta M, Yang T, Chen C, Ding Z . 3D human pose estimation with spatial and temporal transformers. In: Proceedings of the IEEE/CVF International Conference on Computer Vision; 2021 Oct 10-17; Montreal, QC, Canada. Piscataway: IEEE; 2021. p. 11656-65.

[11]	Zhao M, Li T, Alsheikh MA, Tian Y, Zhao H, Torralba A, et al. Through-wall human pose estimation using radio signals. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2018 Jun 18-22; Salt Lake City, UT, USA. Piscataway: IEEE; 2018. p. 7356-65.

[12]	Li G, Zhang Z, Yang H, Pan J, Chen D, Zhang J . Capturing human pose using mmWave radar. In: Proceedings of the IEEE International Conference on Pervasive Computing and Communications Workshops; 2020 Mar 23-27; Austin, TX, USA. Piscataway: IEEE; 2020. p. 1-6.

[13]	Sengupta A, Jin F, Zhang R, Cao S . mm-Pose: real-time human skeletal posture estimation using mmWave radars and CNNs. IEEE Sens J 2020; 20(17): 10032-44.

[14]	Shi C, Lu L, Liu J, Wang Y, Chen Y, Yu J . mPose: environment- and subject-agnostic 3D skeleton posture reconstruction leveraging a single mmWave device. Smart Health 2022; 23: 100228.

[15]	Ho YH, Cheng JH, Kuan SY, Jiang Z, Chai W, Huang HW, et al. RT-Pose: a 4D radar tensor-based 3D human pose estimation and localization benchmark. In: Proceedings of the 18th European Conference on Computer Vision; 2024 Sep 29-Oct 4; Milan, Italy. Berlin: Springer; 2024. p. 107-25.

[16]	Chen L, Wang G . CPFormer: end-to-end multi-person human pose estimation from raw radar cubes with transformers. IEEE Sens J 2025; 25(7): 12466-78.

[17]	Sengupta A, Cao S . mmPose-NLP: a natural language processing approach to precise skeletal pose estimation using mmWave radars. IEEE Trans Neural Netw Learn Syst 2023; 34(11): 8418-29.

[18]	Lai J, Tian J, Zou Y, Song X, Liu F, Li D . mmHPE: human pose estimation based on point cloud from millimeter-wave radar. In: Proceedings of the IEEE International Conference on Systems, Man, and Cybernetics; 2024 Oct 6-10; Kuching, Malaysia. Piscataway: IEEE; 2025. p. 2325-31.

[19]	Shi X, Ohtsuki T . A robust multi-frame mmWave radar point cloud-based human skeleton estimation approach with point cloud reliability assessment. In: Proceedings of the IEEE SENSORS; 2023 Oct 29-Nov 1; Vienna, Austria. Piscataway: IEEE; 2023. p. 1-4.

[20]	Slivko E, Toprak Ö, Raheja V, Bierzynsk K, Servadei L, Wille R . RadarPoseNet: human pose estimation from extremely sparse FMCW radar point clouds. In: Proceedings of the IEEE SENSORS; 2025 Oct 19-22; Vancouver, BC, Canada. Piscataway: IEEE; 2025. p. 1-4.

[21]	He Y, Wang J, Li Y, Luo Y . Dual-path CNN-BiLSTM for mmWave-based human skeletal pose estimation. IEEE Sens J 2025; 25(7): 11683-96.

[22]	Feng Y, Dai S, Zhang Q, Wang Z, Zhang X, Zhou Y . M3Pose: multi-person 3D pose estimation using sparse millimeter-wave radar point clouds. In: Proceedings of the 7th Chinese Conference on Pattern Recognition and Computer Vision; 2024 Oct 18-20; Urumqi, China. Singapore: Springer; 2025. p. 504-17.

[23]	Cao Z, Mei G, Guo X, Wang G . VirTeach: mmWave radar point-cloud-based pose estimation with virtual data as a teacher. IEEE Internet Things J 2024; 11(10): 17615-28.

[24]	Wang Z, Monjur M, Nirjon S . mmJoints: expanding joint representations beyond (x, y, z) in mmWave-based 3D pose estimation. 2025. arXiv:2510.08970.

[25]	Lee SP, Kini NP, Peng WH, Ma CW, Hwang JN . HuPR: a benchmark for human pose estimation using millimeter wave radar. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision; 2023 Jan 2-7; Waikoloa, HI, USA. Piscataway: IEEE; 2023. p. 5704-13.

[26]

Chen A, Wang X, Zhu S, Li Y, Chen J, Ye Q . mmBody benchmark: 3D body reconstruction dataset and analysis for millimeter wave radar. In: Proceedings of the 30th ACM International Conference on Multimedia; 2022 Oct 10-14; Lisboa, Portugal. New York City: Association for Computing Machinery; 2022. p. 3501-10.

[27]

Yang J, Huang H, Zhou Y, Chen X, Xu Y, Yuan S, et al. MM-Fi: multi-modal non-intrusive 4D human dataset for versatile wireless sensing. In: Proceedings of the 37th Conference on Neural Information Processing Systems; 2023 Dec 10-16; New Orleans, LA, USA. Red Hook: Curran Associates, Inc.; 2023. p. 18756-68.

[28]	Su Y, Hou HC, Lan H, Ma CZH . A high-fidelity mmWave radar dataset for privacy-sensitive human pose estimation. Bioengineering 2025; 12(8): 891.

[29]	Fan J, Zhou Y, Yang Y, Cui X, Zhang J, Xie L, et al. M4Human: a large-scale multimodal mmWave radar benchmark for human mesh reconstruction. 2026. arXiv:2512.12378.

[30]	Dong M, Li W, Sun X, Bi S . A global-local network for human pose estimation with completed mmWave radar point clouds. IEEE Sens J 2026; 26(5): 7152-64.

[31]	Yan Z, Ni H, Yu X, Tan W . Indoor 3-D human pose estimation using single millimeter-wave radar and conditional generative adversarial network. IEEE Sens J 2025; 25(13): 23207-15.

[32]

Han Z, Jiang J, Ding X, Wang J, Meng Q, Xu S, et al. DenserRadar: a 4D millimeter-wave radar point cloud detector based on dense liDAR point clouds. In: Proceedings of the IEEE 27th International Conference on Intelligent Transportation Systems; 2024 Sep 24-27; Edmonton, AB, Canada. Piscataway: IEEE; 2024. p. 930-6.

[33]	Yu Y, Chen Z, Zhao C, Liu H, Wu Y, Yin WY, et al. A 39 GHz dual-channel transceiver chipset with an advanced LTCC package for 5G multi-beam MIMO systems. Engineering 2023; 22: 125-40.

[34]	Venon A, Dupuis Y, Vasseur P, Merriaux P . Millimeter wave FMCW radars for perception, recognition and localization in automotive applications: a survey. IEEE Trans Intell Veh 2022; 7(3): 533-55.

[35]	Sun Y, Huang Z, Zhang H, Cao Z, Xu D . 3DRIMR: 3D reconstruction and imaging via mmWave radar based on deep learning. In: Proceedings of the IEEE International Performance, Computing, and Communications Conference; 2021 Oct 29-31; Austin, TX, USA. Piscataway: IEEE; 2021. p. 1-8.

[36]	Bai Y, Liu Z, Duan M, Feng J, Xiong X, Wei K, et al. Microwave photonic links and transversal microwave photonic applications. Electromagn Sci 2025; 3(1): 0090541.

[37]

Singh AD, Sandha SS, Garcia L, Srivastava M . RadHAR: human activity recognition from point clouds generated through a millimeter-wave radar. In: Proceedings of the 3rd ACM Workshop on Millimeter-Wave Networks and Sensing Systems; 2019 Oct 25; Los Cabos, Mexico. Red Hook: Curran Associates, Inc.; 2019.

[38]	Zhao P, Lu CX, Wang B, Trigoni N, Markham A. CubeLearn: end-to-end learning for human motion recognition from raw mmWave radar signals. IEEE Internet Things J 2023; 10(12): 10236-49.

[39]

Mahajan P, Chaudhary D, Khan M, Khan MH, Wajid M, Srivastava A . A point cloud-based non-intrusive approach for human posture classification by utilizing 77 GHz FMCW radar and deep learning models. In: Proceedings of the IEEE International Symposium on Circuits and Systems; 2024 May 19-22; Singapore. Piscataway: IEEE; 2024. p. 1-5.

[40]	Zhu J, Huang X, Deng Z, Qiu Y . mRadHPRS: human pose recognition system from point clouds generated through a millimeter-wave radar. IEEE Trans Aerosp Electron Syst 2025; 61(2): 3227-42.

[41]	Zhang Y. Differential antennas: fundamentals and applications. Electromagn Sci 2023; 1(1): 0010021.

[42]	Cui H, Zhong S, Wu J, Shen Z, Dahnoun N, Zhao Y . MiliPoint: a point cloud dataset for mmWave radar. In: Proceedings of the 37th Conference on Neural Information Processing Systems; 2023 Dec 10-16; New Orleans, LA, USA. Red Hook: Curran Associates, Inc.; 2023. p. 62713-26.

[43]	An S, Li Y, Ogras U. mRI: multi-modal 3D human pose estimation dataset using mmWave, RGB-D, and inertial sensors. In: Proceedings of the 36th Conference on Neural Information Processing Systems; 2022 Nov 28-Dec 9; New Orleans, LA, USA. Red Hook: Curran Associates, Inc.; 2022. p. 27414-26.

[44]	Wei G, Cui C, Dong X. A transformer-based network for human pose estimation using millimeter wave radar data. In: Proceedings of the International Applied Computational Electromagnetics Society Symposium; 2023 Aug 15-18; Hangzhou, China. Piscataway: IEEE; 2023. p. 1-4.

[45]	Zhu B, He Z, Xiong W, Ding G, Huang T, Xiang W . ProbRadarM3F: mmWave radar-based human skeletal pose estimation with probability map-guided multiformat feature fusion. IEEE Trans Aerosp Electron Syst 2025; 61(6): 15832-42.

[46]	Kini NP, Shiue RH, Chandra R, Peng WH, Ma CW, Hwang JN . Transhupr: Cross-view fusion transformer for human pose estimation using mmWave radar. In: Proceedings of the 35th British Machine Vision Conference; 2024 Nov 25-28; Glasgow, UK. Durham: British Machine Vision Association; 2024.

[47]	Qi CR, Su H, Mo K, Guibas LJ . PointNet: deep learning on point sets for 3D classification and segmentation. In: Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition; 2017 Jul 21-26; Honolulu, HI, USA. Piscataway: IEEE; 2017. p. 77-85.

[48]	Qi CR, Yi L, Su H, Guibas LJ . PointNet++: deep hierarchical feature learning on point sets in a metric space. In: Proceedings of the 31st International Conference on Neural Information Processing Systems; 2017 Dec 3-9; Long Beach, CA, USA. Red Hook: Curran Associates, Inc.; 2017. p. 5105-5114.

[49]	Yu X, Rao Y, Wang Z, Liu Z, Lu J, Zhou J . PoinTr: diverse point cloud completion with geometry-aware transformers. In: Proceedings of the IEEE/CVF International Conference on Computer Vision; 2021 Oct 10-17; Montreal, QC, Canada. Piscataway: IEEE; 2021. p. 12498-507.

[50]	Wu Y, Jiang Z, Ni H, Mao C, Zhou Z, Wang W, et al. mmHPE: robust multiscale 3-D human pose estimation using a single mmwave radar. IEEE Internet Things J 2025; 12(1): 1032-46.

[51]	Su Y, Hou HC, Zhao C, Nan Z . PoseGraphNet: pose prior and graph structure for 3D human pose estimation using mmWave radar. Measurement 2026; 257: 118851.

[52]	Hu S, Cao S, Toosizadeh N, Barton J, Hector MG, Fain MJ . mmPose-FK: a forward kinematics approach to dynamic skeletal pose estimation using mmWave radars. IEEE Sens J 2024; 24(5): 6469-81.

[53]	Engel L, Mueller J, Rendon EJF, Dorschky E, Krauss D, Ullmann I, et al. Advanced millimeter wave radar-based human pose estimation enabled by a deep learning neural network trained with optical motion capture ground truth data. IEEE J Microwaves 2025; 5(2): 373-87.

[54]	Luo Y, He Y, Li Y, Liu H, Wang J, Gao F . A sliding window-based CNN-BiGRU approach for human skeletal pose estimation using mmWave radar. Sensors 2025; 25(4): 1070.

[55]	Lu W, Chen Y, Huang X . mmChainPose: geometry-aware temporal chaining for robust human pose estimation from mmWave point clouds. Neurocomputing 2026; 663: 132019.

[56]

Kini NP, Tsai SR, Lin GH, Peng WH, Ma CW, Hwang JN . milliMamba: specular-aware human pose estimation via dual mmWave radar with multi-frame mamba fusion. In: Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision; 2026 Mar 6-10; Tucson, AZ, USA. Piscataway: IEEE; 2026. p. 1481-90.

[57]	Ma N, Pan J, Liu Y, Yang Y, Han Y, Guo J, et al. Embodied interactive intelligence towards autonomous driving. Engineering 2026; 59: 337-51.

[58]	An S, Ogras UY . MARS: mmWave-based assistive rehabilitation system for smart healthcare. ACM Trans Embed Comput Syst 2021; 20(5s): 72.

[59]	Roshandel N, Scholz C, Cao HL, Cao HG, Amighi M, Firouzipouyaei H, et al. mmPrivPose3D: a radar-based approach to privacy-compliant pose estimation and gesture command recognition in human-robot collaboration. IEEE Sens J 2025; 25(15): 29437-45.

[60]	Zhao H, Jiang L, Jia J, Torr PH, Koltun V . Point Transformer. In: Proceedings of the IEEE/CVF International Conference on Computer Vision; 2021 Oct 11-17; online. Piscataway: IEEE; 2021. p. 16259-68.

[61]	Ma X, Qin C, You H, Ran H, Fu Y . Rethinking network design and local geometry in point cloud: a simple residual MLP framework. 2022. arXiv:2202.07123.

[62]	Wang Y, Sun Y, Liu Z, Sarma SE, Bronstein MM, Solomon JM . Dynamic graph CNN for learning on point clouds. ACM Trans Graph 2019; 38(5): 146.

[63]	Otte K, Kayser B, Mansow-Model S, Verrel J, Paul F, Brandt AU, et al. Accuracy and reliability of the kinect version 2 for clinical measurement of motor function. PLoS One 2016; 11(11): e0166532.

[64]	Faity G, Mottet D, Froger J . Validity and reliability of Kinect V2 for quantifying upper body kinematics during seated reaching. Sensors 2022; 22(7): 2735.

[65]	Ogawa A, Mita A, Yorozu A, Takahashi M . Markerless knee joint position measurement using depth data during stair walking. Sensors 2017; 17(11): 2698.