Recognizing Actions from Depth Cameras as Weakly Aligned Multi-part Bag-of-Poses

Recently released depth cameras provide effective esti-mation of 3D positions of skeletal joints in temporal se-quences of depth maps. In this work, we propose an effi-cient yet effective method to recognize human actions based on the positions of joints. First, the body skeleton is de-composed in a set of kinematic chains, and the position of each joint is expressed in a locally defined reference sys-tem which makes the coordinates invariant to body trans-lations and rotations. A multi-part bag-of-poses approach is then defined, which permits the separate alignment of body parts through a nearest-neighbor classification. Ex-periments conducted on the Florence 3D Action dataset and the MSR Daily Activity dataset show promising results. 1.