Lightweight Human Posture Estimation Incorporating Hybrid Attention Mechanisms

Aiming at the problem that the existing human posture estimation models have high prediction accuracy and large computational complexity at the same time, this paper innovatively proposes LCBAM (Lightweight Convolutional Block Attention Module), which incorporates the hybrid attention mechanism, on the basis of the HRNet model. The new model adopts the Lightweight Design Sampling Module to reduce the running complexity of the model, and at the same time introduces the Hybrid Attention Mechanism Module (CBAM) to adaptively capture global features, so as to satisfy the lightweight design goal and improve the estimation accuracy of the model at the same time. The experimental results show that, compared with the benchmark model, the model in this paper exhibits obvious advantages in the detection of key points for posture estimation, especially in the detection of key points for the left wrist, right wrist, left knee, right knee, left ankle, and right ankle, with enhancements of 5.2%, 5%, 2.5%, 3.41%, 2.8%, and 4.04%, respectively. At the same time, the design of the fusion hybrid attention mechanism does not significantly increase the computational complexity of the lightweight network, but rather improves the accuracy of the key point detection, which solves the contradiction between the lightweight and prediction accuracy of the pose estimation model to a certain extent.