The invention discloses a method for inferring a human body action pose through a 2D image. The method comprises the following steps: a 2D image obtaining step of obtaining the 2D image containing a human body action, a confidence map calculation step of performing a convolutional neural network algorithm on the 2D image to obtain a corresponding confidence map, a matching step of matching the confidence map corresponding to the 2D image with a preset 3D model database to obtain a 3D model corresponding to the 2D image, and a mapping step of performing mapping on the 3D model under any plane angle to obtain the human body action pose under each plane angle. The invention furthermore provides an electronic device and a computer readable storage medium. The problem of low identification rateof the human body action pose obtained through the 2D image in the prior art is solved.