Learning the hand skeleton from 2D images, lifting and denoising

摘要

The processor identifies keypoints 115 , 120 , 125 , 130 of the hand in the two-dimensional image 100 captured by the camera 215 . A three-dimensional pose of the hand is determined using the location of the keypoint to access a lookup table (LUT) 230 that represents the potential pose of the hand as a function of the location of the keypoint. In some embodiments, the keypoint is the locations of the tips of the fingers and thumb, the joints connecting the fingers and the phalanges of the thumb, palm knuckles indicating the point of attachment of the fingers and thumb to the palm, and the hand on the forearm. It includes a wrist position indicating the point of attachment. Some embodiments of the LUT represent the 2D coordinates of the fingers and thumb in the corresponding finger pose plane.