Human body action recognition method, system, equipment and medium

The invention discloses a human body action recognition method, system and device and a medium, and relates to the technical field of computer vision, and the method comprises the steps: replacing a Focus layer of an input end in a posture estimation model YOLOPose with a convolution layer with a convolution kernel of 6 * 6 according to a human body action video data set, deleting 64-time down-sampling detection heads in the detection heads and network layers related to the 64-time down-sampling detection heads; a double-attention mechanism is added to the YOLOPose backbone network; a PAN structure in the YOLOPose detection head is replaced by a bidirectional feature pyramid, so that an improved YOLOPose model is formed; the improved YOLOPose model is trained through a human body action video data set; inputting a to-be-detected human body action video data set into the trained YOLOPose model to obtain human body skeleton node coordinates; and inputting the human skeleton node coordinates into a space-time di