Human motion evaluation method based on two-dimensional skeleton sequences

Abstract

Provided is a human motion evaluation method based on two-dimensional skeleton sequences. The method includes the steps that an RGB image containing a human body is acquired through a camera, and a two-dimensional human skeleton is extracted, wherein the human skeleton has several joint points; the initial pose of the human skeleton is corrected; four features including joint point coordinates, key part angles, joint point velocity and fluctuation errors of the joint point velocity are extracted to form a feature matrix, and a corresponding frame difference degree calculation model is constructed; matrix coordinate systems of a testing skeleton sequence and a standard skeleton sequence are established, front and rear segments of the two sequences are searched to obtain matching path starting and ending points, global bidirectional parallel searching is conducted with the matching path starting and ending points as reference points until optimal path matching of the testing skeleton sequence and the standard skeleton sequence is completed, and the accumulative difference value of the two sequences is obtained to realize human motion evaluation. According to the method, the low-costRGB camera is adopted to conduct accurate evaluation and trend analysis on the human motion, and the method has high application value in motion training and guidance of sports, martial arts, dances and the like.

Description

technical field [0001] The invention relates to a human body motion evaluation method based on a two-dimensional skeleton sequence. Background technique [0002] With the progress of human society and the development needs of people's lives, artificial intelligence technologies such as intelligent visual recognition, natural human-computer interaction, and virtual augmented reality are gradually and widely used in people's lives. Among them, articulated objects ( Such as the detection, matching and recognition of the posture of the human body is an important research content in the field of intelligent vision and human-computer interaction, and can be widely used in medical rehabilitation training, sports training action analysis and evaluation, virtual reality human-computer interaction and other fields . [0003] Traditional motion analysis methods for articulated objects include contact methods such as wearing magnetic trackers and optical tags on the human body or using...

AI Technical Summary

Problems solved by technology

In addition, although depth cameras such as Microsoft Kinect and Intel Realsense can collect color and depth images and have advantages in extracting human skeleton models, they are not suitable for existing surveillance systems due to problems such as high cost and low resolution. equipment, unable to meet the growing needs of ordinary users in sports evaluation, human-computer interaction, virtual reality, etc., which is not conducive to popularization and application

Images