Abnormal state detection method for digital human gesture action

The invention provides an abnormal state detection method for gesture actions of a digital human, which comprises the following steps: a trained reconstruction model obtained according to the implementation method of the invention is obtained, and training samples adopted during training are all gesture action states of the digital human without a hand mold penetrating condition; a synthetic gesture action sequence of the digital person needing to be detected is obtained, the gesture action state synthesized by each frame is input into the trained reconstruction model, the reconstructed gesture action state is obtained, and the gesture action state comprises the spatial positions of multiple joints of the digital person at one time point; and determining whether the synthesized gesture action state of each frame is abnormal or not according to the distance between the synthesized gesture action state of each frame and the reconstructed gesture action state corresponding to the synthesized gesture action state