Aiming at the problem of object frame positioning error and redundant detection in the top-down human pose estimation algorithm, a top-down human pose estimation algorithm based on deep learning is proposed. The symmetric space transformation network is designed to connect with the single-person pose estimation network to propose high-quality human target frames from inaccurate human body bounding boxes, and parametric pose non-maximum suppression is introduced to eliminate redundant pose estimation , The elimination rule is applied to eliminate similar postures, and the unique human posture estimation result is obtained. Part of the data set is selected for training and testing on the public human pose estimation data set MPII. The experimental results show that the method proposed in this paper can accurately detect the key points of the human body, effectively improve the accuracy of human body pose estimation, and can adapt to crowded people. , Complex scenes with occlusion.