Title:
METHOD FOR HUMAN ACTION RECOGNITION IN HUMAN-MACHINE INTERACTIVE ASSEMBLY SCENE
Document Type and Number:
Japanese Patent JP2023108595
Kind Code:
A
Abstract:
To provide a method for human action recognition in a human-machine interactive assembly scene.SOLUTION: The method for human action recognition comprises the steps of: obtaining joint coordinate streams of skeleton joints under a human action from motion sensing devices; screening a joint coordinate stream with complete skeleton joints; determining a starting position and an ending position of the action on the basis of an action event segmentation algorithm to obtain information of joints; performing resampling of angle change on the information of joints in accordance with an included angle θ to obtain coordinates of joints; normalizing and smoothing the coordinates of the other joints to obtain a sequence of skeletons forming one action; simplifying a vector of adjacent joints of upper limbs to obtain a vector direction of the upper limbs; calculating included angles between vector directions of a left upper limb and a right upper limb and a vertical direction; and dividing a scene to a left-hand scene and a right-hand scene according to the included angles; performing training for human action recognition in the scenes respectively; and fusing human action outputs of the scenes to perform action recognition in a human-machine interaction scene.SELECTED DRAWING: Figure 1
Inventors:
CHEN PENGZHAN
LI FANG
LI FANG
Application Number:
JP2022193633A
Publication Date:
August 04, 2023
Filing Date:
December 02, 2022
Export Citation:
Assignee:
UNIV TAIZHOU
International Classes:
G06T7/00; G06T7/20; G06T7/70; G06V40/20
Attorney, Agent or Firm:
Koji Tsubouchi