PoseRL-Net: Human Pose Analysis for Motion Training Guided by Robot Vision

Bin Liu Hui Wang ^*

• Shanghai Jianqiao University, Shanghai, Shanghai, China

In the field of collaborative robotics, precise human pose recognition is a key technology for achieving seamless human-robot interaction, especially in complex dynamic environments.Traditional methods face limitations in handling occlusions, lighting variations, and motion continuity. To address these challenges, we propose a deep learning-based pose recognition model, PoseRL-Net, designed to enhance accuracy and robustness in human pose estimation.PoseRL-Net integrates components such as a Spatial-Temporal Graph Convolutional Network (STGCN), attention mechanism, Gated Recurrent Unit(GRU) module, pose refinement, and symmetry constraints to build a robust pose estimation framework. The STGCN effectively extracts spatial and temporal features, the attention mechanism improves focus on key pose features, the GRU module ensures temporal consistency in actions, and the pose refinement and symmetry constraints enhance structural plausibility and stability in predictions. Extensive experiments on the Human3.6M and MPI-INF-3DHP datasets demonstrate that PoseRL-Net surpasses existing state-of-the-art models on key metrics such as MPJPE and P-MPJPE, highlighting its superior performance across various pose recognition tasks. PoseRL-Net not only improves pose estimation accuracy but also provides crucial support for intelligent decision-making and motion planning for robots in dynamic and complex scenarios, offering significant practical value.