Improved Human Pose Estimation Algorithm Based on HigherHRNet

The problem of imbalance between foreground and background samples exists in the bottom-up human pose estimation method, while the high-resolution network cannot effectively obtain the channel information and spatial position information during feature extraction and feature fusion. To address the above problems, this paper proposes a bottom-up human pose estimation network based on the high-resolution network HigherHRNet and fused with pyramid network. The pyramid network can provide multi-scale feature representation according to different levels of sensory field size, extract contextual information from different levels of features as well as fusion between different levels of features to improve the feature expression capability. Combining the pyramid network with HigherHRNet can improve the performance of small human target detection. The detection and localization of small human targets can be effectively enhanced by multi-scale sensory fields, rich contextual information, feature fusion and scale adaptation, and the accuracy and robustness of the algorithm can be improved.