A Lightweight Human Pose Estimation Algorithm Based on High Resolution Network

Human pose estimation is an important research direction in the field of computer vision. At present, the mainstream human pose estimation algorithms have high complexity, large amount of calculation, and cannot be run on resource-constrained devices such as mobile terminals, which severely limits the popularization and application of this technology. Aiming at the problem of increased network model parameters and computational complexity, based on the High-Resolution Network (HRNet), a lightweight human pose estimation network incorporating Ghost module and attention mechanism is proposed. Replaced with Ghost convolution, and added the attention mechanism Concurrent Spatial and Channel Squeeze and Channel Excitation Net module on this basis to ensure the prediction accuracy of the network. Under the same image resolution and environment configuration, the experimental results on the COCO dataset show that the improved network model reduces the number of parameters by 98.3% compared to the high-resolution network model, and reduces the computational complexity by 67.6%. The experimental results show that the improved network model can effectively reduce the amount of network parameters and reduce the computational complexity while maintaining a certain prediction accuracy.