3D human body posture estimation method based on space-time cycle architecture

According to the 3D human body posture estimation method based on the space-time cycle architecture, three image convolution networks are designed respectively to process image data, a fused feature is finally obtained, spatial correlation between joints is fully learned, and meanwhile the problem of self-shielding of the joints is solved to a certain degree; a global and local alternating mechanism is adopted, and a self-attention mechanism which is excellent in modeling global information and a sliding window mechanism which can control the attention range of the attention mechanism are combined, so that a network is guided to pay attention to local details in a short time and pay attention to global information in a long time; therefore, the method successfully avoids the interference of a longer sequence, enables the improved self-attention to alternately pay attention to global and local information, more comprehensively captures time features, and solves the problem of depth ambiguity of joints to a cer