MOTION MODEL LEARNING DEVICE, THREE-DIMENSIONAL ATTITUDE ESTIMATION DEVICE, MOTION MODEL LEARNING METHOD, THREE-DIMENSIONAL ATTITUDE ESTIMATION METHOD AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To improve the accuracy of attitude estimation without manually designating joint positions.

SOLUTION: In learning, in addition to construction of one latent variable model M from information on joints of the whole body, a latent variable model M1 is learned from information on selection points, the selection points being partial joints of the whole body, and a motion model for mapping the latent variable model M to the latent variable model M1 is obtained by learning. This enables a three-dimensional motion information processing device 100 to easily perform image recognition on a two-dimensional image with reference to the motion model in estimation, and to obtain the positions of all joints in a latent space of the latent variable model M based on motions of partial joints to be easily tracked.

COPYRIGHT: (C)2012,JPO&INPIT