Research on Human Motion Recognition Methods Based on MediaPipe Pose
To surmount the prevalent challenges of low recognition efficiency and sluggish detection velocities in extant human action recognition algorithms,a novel approach predicated on the MediaPipe Pose algorithm has been developed.This approach involves the real-time acquisition of data via a camera,which is processed by a detection network to ascertain the coordinates of thirty-three salient body keypoints.Subsequently,the configuration of these keypoints in space is utilized to categorize human actions.Action categories are delineated according to the COCO dataset nomenclature,with action labels subjected to one-hot encoding.This procedure facilitates the refinement of a model dedicated to human action recognition.Experimental validation was conducted on eight distinct actions using a monocular RGB camera.Findings indicate that the action recognition method predicated on the MediaPipe Pose algorithm attained a frame rate of 30 f/s and a precision rate of 96.67%,thereby ensuring the real-time and accurate discernment of human movements.
万方数据
维普
