A human body action standard recognition method and system based on deep learning

By constructing a spatiotemporal sequence and fusion features of key points of the human skeleton, and combining body morphology parameters and multi-dimensional difference measurement, the problem of subjectivity and detail capture in human motion recognition is solved, and personalized and refined motion evaluation is achieved.

CN122200818APending Publication Date: 2026-06-12THE SECOND HOSPITAL AFFILIATED TO WENZHOU MEDICAL COLLEGE
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
THE SECOND HOSPITAL AFFILIATED TO WENZHOU MEDICAL COLLEGE
Filing Date
2026-05-15
Publication Date
2026-06-12

AI Technical Summary

⚠Technical Problem

Existing technologies struggle to achieve refined, personalized, and objective evaluations in human motion recognition. They are heavily influenced by the observer's subjective factors and cannot adequately capture the details of multiple people's movements or subtle deviations.

⚗Method used

By constructing a spatiotemporal sequence of key points of the human skeleton, integrating action posture features and temporal features, generating action representation vectors, and introducing body morphology parameters for personalized adaptation, the system utilizes a multi-dimensional difference measurement branch for evaluation, and combines a feature recalibration module to improve discrimination capabilities.

🎯Benefits of technology

It enables refined, personalized, and objective evaluation of human movements, improves the accuracy and fairness of evaluations, and provides automated, standardized assessment tools.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122200818A_ABST
    Figure CN122200818A_ABST
Patent Text Reader

Abstract

The application relates to the technical field of human action recognition, and discloses a human action standardization recognition method and system based on deep learning, which comprises the following steps: collecting continuous multiple frames of human images of a target object performing a preset action; inputting the images into a first deep learning network, extracting human skeleton key point coordinates, arranging the human skeleton key point coordinates according to time sequences, and constructing a human skeleton key point space-time sequence; inputting the space-time sequence into a second deep learning network, extracting action posture features and action time sequence features, fusing the action posture features and the action time sequence features, and generating an action representation vector; obtaining standard action parameters corresponding to the preset action; inputting the action representation vector and the standard action parameters into a comparison model, analyzing a difference degree, and outputting a standardization score result. The application fuses posture features and time sequence features by constructing a space-time sequence, and realizes individual evaluation in combination with body shape adaptation, thereby improving the accuracy and objectivity of action standardization recognition.
Need to check novelty before this filing date? Find Prior Art