A method and system for twin pose detection based on interactive indirect inference. The method includes: acquiring, by sensors on a mobile phone, a data set in real time; and obtaining poses of individual parts of a skeleton of an object by reasoning using reasoners on individual parts based on the data set and a preferred way of the object for using the mobile phone; merging the poses to generate multiple initial virtual skeletons; under a predetermined human mechanics constraint, obtaining multiple overall virtual skeletons satisfying the predetermined human mechanics constraint from the initial virtual skeletons; and screening a predetermined number of overall virtual skeletons from the overall virtual skeletons, and obtaining a dynamic twin virtual pose in real time by reasoning on the predetermined number of overall virtual skeletons using an overall skeleton reasoner.