Workshop Worker Pose Estimation Method Based on Spatio-Temporal Cross-Attention

This paper propose an STC (Spatio-Temporal Cross-Attention) based approach for estimating worker postures on the shop floor, aiming to address the challenge of accurate estimation of worker postures in industrial environments. In industrial scenarios on the manufacturing floor, accurately capturing and understanding workers’ postures is crucial for optimizing production processes, improving work safety, and increasing efficiency. However, due to the complexity and dynamics of industrial environments, traditional pose estimation methods often face challenges. In this paper, we design STCFormer by stacking multiple STC blocks to establish correspondence in space and time to effectively track workers’ postures. STC can accurately match workers’ critical joints in different periods and camera views, thus realizing continuous monitoring and estimation of workers’ postures. Compared with traditional static image-based methods, STCFormer is more suitable for dynamic industrial environments and can improve the accuracy and stability of posture estimation. To verify the effectiveness of the proposed method, a series of experiments are conducted in the real industrial workshop environment in this paper. The experimental results show that the STC-based pose estimation method can achieve significant improvements in the accuracy and stability of estimating different movements and poses of workers in complex industrial scenarios.