A motion quality evaluation method and system

Abstract

The invention discloses a motion quality evaluation method and system, wherein the method comprises extracting a local motion pattern of each part of that body from the motion trajectory of a human joint, and establishing the behavior classification to distinguish the movement pattern, and then according to the established behavior classification for judging the motion mode, establishing a model of pose feature representation and quality evaluation based on human joint displacement correlation, and then scoring the quality of human motion in the video captured by the camera according to the pose feature representation and quality evaluation model based on human joint displacement correlation. By using the method, it is possible to accurately feedback the evaluation information of the humanmotion quality without labeling the human motion data manually.

Description

technical field [0001] The invention relates to the technical field of motion quality evaluation, in particular to a motion quality evaluation method and system. Background technique [0002] In terms of the sports industry, sports quality evaluation technology can enable athletes to obtain real-time scores on the quality of their movements while training under cameras, and visualized feedback data can help athletes improve their performance. The sports assistance system based on this technology can detect wrong postures to prevent injuries caused by sports by adding physiological constraints of human sports. Therefore, it has broad application prospects in related fields such as competitive sports training, sports teaching and sports rehabilitation. In the field of content-based video retrieval, search engines can not only return videos related to query conditions based on similarity metrics, but also sort the retrieved videos with the assistance of motion quality evaluatio...

AI Technical Summary

Problems solved by technology

[0005] However, the existing motion quality evaluation scheme integrates the feature representation and regression prediction of all motion categories into one evaluation function for optimization, so the verification process only targets one or two categories of motion data, and requires a specific domain when adding motion categories of experts label the data

In addition, when there are large differences in the motion patterns of different categories, using a single evaluation function for evaluation is prone to be affected by the distribution of sample data and cause large approximation errors.

Moreover, similar postures of different behavior categories may be confused, causing large deviations or even unreasonable feedback information on the position feedback of joint points.

Images