The present disclosure relates to a method for automatically generating labeled data of a hand, comprising: acquiring at least three images to be processed of the hand under different angles of view; detecting key points on the at least three images to be processed respectively; screening the detected key points by using an association relation among the at least three images to be processed, the association relation being the same frame of image of the at least three images to be processed from the hand under different angles of view; reconstructing a three-dimensional space representation of the hand with regard to the key points screened on the same frame of image, in combination with a given finger bone length; projecting the key points on the three-dimensional representation of the hand onto the at least three images to be processed; and generating the labeled data of the hand on the images to be processed by using the projected key points on the at least three images to be processed.