Song Liang
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Song Liang has a total of 7 co-authors. Here are the top co-authors by number of shared publications:
- Jun Wang
School of Electronic and Information Engineering, Beihang University, Beijing 100191, China
2 shared publications - Yaxin Li
Hangzhou Innovation Institute, Beihang University, Hangzhou, China
2 shared publications - Yuquan Luo
School of Electronic and Information Engineering, Beihang University, Beijing, China
2 shared publications Changshun Yuan
Hangzhou Innovation Institute, Beihang University, Hangzhou, China
1 shared publication- Changshun Yuan
Hangzhou Innovation Institute, Beihang University, Hangzhou 310051, China
1 shared publication Xiangqian Li
School of Electronic and Information Engineering, Beihang University, Beijing, China
1 shared publication
Latest Publications
- 22 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (14) , 31463-31478
Multiscale Spatial–Temporal Representation Learning for mmWave Radar 3-D Human Pose Estimation
Millimeter-wave (mmWave) radar has emerged as a promising sensing modality for 3D human pose estimation in ubiquitous Internet of Things (IoT) applications, owing to its privacy-preserving nature and robustness under challenging illumination conditions. However, accurate skeletal reconstruction remains difficult due to the inherent sparsity, non-uniform distribution, and instability of radar point clouds. To address these challenges, this paper proposes MS-STPoseNet, a unified multi-scale spatio-temporal learning framework for robust mmWavebased human pose estimation. For spatial representation, MS-STPoseNet employs a hierarchical multi-scale spatial encoder built upon PointNet++, which leverages multi-scale grouping to effectively capture body structures across different spatial resolutions, from fine-grained joint regions to limb- and torso-level configurations, under irregular radar observations. For temporal modeling, a multi-branch Temporal Convolutional Network (TCN) with different dilation rates is introduced to model multi-rate motion dynamics, enabling effective representation of both rapid limb movements and smoother torso motions. An attention mechanism is further incorporated to enhance informative temporal features while suppressing noise. The entire framework is trained end-to-end to estimate per-frame 3D poses by exploiting short-term temporal context, thereby improving robustness under noisy sensing conditions. Extensive experiments conducted on a self-collected dataset and two public benchmarks demonstrate that MS-STPoseNet consistently outperforms state-of-the-art methods in terms of pose estimation accuracy and cross-subject generalization, achieving an MPJPE of 3.08 cm on the self-collected dataset. In addition, the proposed framework exhibits favorable computational efficiency and a compact model size, highlighting its potential applicability to practical IoT sensing systems.
- 22 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Internet of Things Journal
- Vol. 13 (14) , 31463-31478
Multiscale Spatial–Temporal Representation Learning for mmWave Radar 3-D Human Pose Estimation
Millimeter-wave (mmWave) radar has emerged as a promising sensing modality for 3D human pose estimation in ubiquitous Internet of Things (IoT) applications, owing to its privacy-preserving nature and robustness under challenging illumination conditions. However, accurate skeletal reconstruction remains difficult due to the inherent sparsity, non-uniform distribution, and instability of radar point clouds. To address these challenges, this paper proposes MS-STPoseNet, a unified multi-scale spatio-temporal learning framework for robust mmWavebased human pose estimation. For spatial representation, MS-STPoseNet employs a hierarchical multi-scale spatial encoder built upon PointNet++, which leverages multi-scale grouping to effectively capture body structures across different spatial resolutions, from fine-grained joint regions to limb- and torso-level configurations, under irregular radar observations. For temporal modeling, a multi-branch Temporal Convolutional Network (TCN) with different dilation rates is introduced to model multi-rate motion dynamics, enabling effective representation of both rapid limb movements and smoother torso motions. An attention mechanism is further incorporated to enhance informative temporal features while suppressing noise. The entire framework is trained end-to-end to estimate per-frame 3D poses by exploiting short-term temporal context, thereby improving robustness under noisy sensing conditions. Extensive experiments conducted on a self-collected dataset and two public benchmarks demonstrate that MS-STPoseNet consistently outperforms state-of-the-art methods in terms of pose estimation accuracy and cross-subject generalization, achieving an MPJPE of 3.08 cm on the self-collected dataset. In addition, the proposed framework exhibits favorable computational efficiency and a compact model size, highlighting its potential applicability to practical IoT sensing systems.
- 30 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (24) , 44363-44377
A Spatiotemporal Attention Network for mmWave-Based 3-D Human Skeleton Estimation
This article presents a spatiotemporal attention framework for radar-based human pose estimation (STAR-Pose), a unified model for estimating 3-D skeletal coordinates directly from millimeter-wave (mmWave) radar point clouds. The framework integrates PointNet++ for spatial feature extraction and a bidirectional long short-term memory (BiLSTM) module enhanced by attention pooling for temporal modeling. Data were collected using a selfdeveloped BHYY_MMW6044 radar operating in the 59–64-GHz band, capturing dynamic human motion in diverse scenarios. Comprehensive experiments demonstrate that STAR-Pose achieves an average localization error of 1.76 cm, outperforming existing radar-based baselines. The framework exhibits strong robustness to noisy frames, varying motion speeds, and cross-subject conditions, while maintaining stable accuracy under occlusion and multipath interference. Overall, STAR-Pose provides a reliable and privacy-preserving approach for human pose estimation with mmWave radar, paving the way for intelligent sensing applications in smart healthcare, ambient monitoring, and human–computer interaction.
- 30 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Sensors Journal
- Vol. 25 (24) , 44363-44377
A Spatiotemporal Attention Network for mmWave-Based 3-D Human Skeleton Estimation
This article presents a spatiotemporal attention framework for radar-based human pose estimation (STAR-Pose), a unified model for estimating 3-D skeletal coordinates directly from millimeter-wave (mmWave) radar point clouds. The framework integrates PointNet++ for spatial feature extraction and a bidirectional long short-term memory (BiLSTM) module enhanced by attention pooling for temporal modeling. Data were collected using a selfdeveloped BHYY_MMW6044 radar operating in the 59–64-GHz band, capturing dynamic human motion in diverse scenarios. Comprehensive experiments demonstrate that STAR-Pose achieves an average localization error of 1.76 cm, outperforming existing radar-based baselines. The framework exhibits strong robustness to noisy frames, varying motion speeds, and cross-subject conditions, while maintaining stable accuracy under occlusion and multipath interference. Overall, STAR-Pose provides a reliable and privacy-preserving approach for human pose estimation with mmWave radar, paving the way for intelligent sensing applications in smart healthcare, ambient monitoring, and human–computer interaction.