人体关键点捕捉下上肢运动姿态监测

打开文本图片集

中图分类号：TN911.73-34；TP391 文献标识码：A 文章编号：1004-373X（2025）15-0165-04

Upper limb motion posture monitoring based on capture of key points of human body

BOHuimin，LI Huijun （InnerMongolia University，HohhotO1oo22，China）

Abstract：Theamplitudeandspeedof diferent motionsvary，which leads to limbocclusionand furthermore fects the accuracyof motionanalysis，soamethodofupperlimbmotionposture monitoringbasedoncaptureofkeypointsof humanbody is proposed inorder to handle theupper limb motion paterns with temporal dynamic characteristics.The Kinect v2 wasused to obtainthedepth image informationofmotionsof thehuman bodies.Thebackground diference methodwasused toseparate the target mageof the humanupperlimbsand inputit intoDFCRF-Net.BycombiningGCNand CNN，theatention gate mechanism andtheCRF（conditionalrandom field）reasoning wereused to extractthe keypoints of the humanupper limbs.Afterconverting thecoordinatesequenceofbonepointsintoangulartimeseries，DTWalgorithmwasusedtomatchthesimilaritybetweenreal-time upper limb motionsand standardtemplates.The threshold value was settojudge whether themotionposture was standardornot， andtheonitoringofupperlimbmotionposturewasrealized.Theexperimentalresultsshowthatthepositioningaccuracyof the proposed method in the key points of shoulder，elbow and wrist reaches 98.2%，96.5% and 94.7% ，respectively，and the monitoringresultsarehighlyconsistentwiththelawofhumankinematicswhichverifytheefectivenessof theproposedmethod.

Keywords：key point of human body；Kinect v2 ；Sobel operator;background diference method;DFCRF-Net；DTW algorithm; upper limb motion; attitude monitoring

0 引言

上肢作为人体执行精细操作与复杂交互的核心部位，其运动涉及肩、肘、腕及手指等多关节协同作用，具有自由度高、动作灵活性强、空间轨迹复杂等特点。（剩余5390字）