关键词: 计算机视觉, 人体姿态估计, 多重分离沙漏模块, 混洗高效通道注意力, 空间通道感知模块, 冗余特征抑制, 多尺度特征融合

Abstract: A lightweight human pose estimation network based on redundant feature suppression was proposed to address the difficulty of balancing computational efficiency and localization accuracy in complex scenarios. It was termed LE-SHNet(Lightweight Enhanced Stacked Hourglass Network). First, the Multiple Separated Hourglass Module (MSHM) was designed to employ heterogeneous convolution branches for differential modeling of large joints and distal limbs, while suppressing redundant computations. Then, the Shuffle Efficient Channel Attention (SECA) was integrated between hourglass modules, which combines channel shuffling and adaptive kernel convolution to enhance long-range joint correlations with zero additional parameters. Finally, the Spatial and Channel Perception Module (SCPM) was constructed in non-hourglass modules to strengthen spatial attention and channel responses by introducing spatial-channel reconstruction and triplet attention. Experimental results show that LE-SHNet achieves accuracy scores of 88.7% on Max Planck Institute for Informatics （MPII）and 71.3% on Common Objects in COntext 2017(COCO2017), reducing parameters by 49.3% and computational cost by 28.2% compared with the baseline 2 Stacked Hourglass Network (2-SHNet). Compared with the lightweight human pose estimation networks proposed in 2024 and 2025, namely EL-HRNet (Efficient and Lightweight High-Resolution Network) and MobileMultiPose (Mobile-friendly and Multi-feature aggregation Pose estimation), LE-SHNet achieves accuracy improvements of 1.0 and 0.8 percentage points, while reducing the number of parameters by 32.0% and 26.7%, respectively. These findings indicate that LE-SHNet maintains lightweight properties while significantly improving keypoint localization accuracy, making it suitable for real-time deployment on edge devices with promising applications in intelligent surveillance, human–computer interaction, and rehabilitation monitoring.

Key words: computer vision, human pose estimation, Multiple Separated Hourglass Module(MSHM), Shuffle Efficient Channel Attention(SECA), Spatial and Channel Perception Module(SCPM), redundant feature suppression, Multi-scale feature fusion