基于Mediapipe框架的动作识别研究  

Research of Motion Recognition Based on Mediapipe Framework

作　　者：亢越 张建芬[1] 葛世辰 Kang Yue;Zhang Jian-fen;Ge Shi-chen(International School of Energy,Jinan University,Zhuhai 519070,Guangdong Province,China)

机构地区：[1]暨南大学国际能源学院,广东珠海519070

出　　处：《科学与信息化》2025年第1期31-33,共3页Technology and Information

摘　　要：针对办公人群长时间久坐与使用电脑造成的健康问题,本文设计用于康复训练的动作识别方法。该方法依据Mediapipe框架提取人体关节点坐标,计算与关节点相关的距离、角度等几何参数约束条件,同时考虑动作检测的时间约束条件,实现了训练动作的识别。实验结果表明,所提出的动作识别方法具有较高的识别准确度、实时性好,同时具备了计算资源少、轻量化、便于移植到移动终端的特点。Aiming at the health problems caused by long-time sitting and using computers among office people,this paper designs a motion recognition method for rehabilitation training.The method extracts the coordinates of human joint points byusingMediapipe framework,calculates the geometric parameter constraints of distances and angles,and considers the time constraints of motion detection,so as to realize the recognition of training motions.The experimental results show that the proposed motion recognition method has high recognition accuracy and good real-time performance.At the same time,it has the characteristics of less computing resources,lightweight and easy to transplant to mobile terminals.

关 键 词：Mediapipe 运动识别 格拉布斯检测 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]