计算机科学
人工智能
适应性
预警系统
图形
人机交互
机器学习
运动(音乐)
一般化
实时计算
人工神经网络
模拟
控制(管理)
卷积神经网络
计算机视觉
匹配(统计)
可视化
运动(物理)
警报
软件
支持向量机
抓住
预警系统
可视化程序设计语言
动作(物理)
数据收集
毒物控制
深度学习
控制流程图
可用性
步态
运动捕捉
运动控制
假警报
作者
Lida Jin,Chunhai Zhang,Shenbo Lei
标识
DOI:10.1142/s0219519426500314
摘要
To address core pain points in real-time, accurate evaluation of movement posture, lack of personalized guidance, and insufficient early warning of injury risk in high school physical education, this study develops Smart Physical Observation and Safety Education (SPOSE). SPOSE deeply fuses Red Green Blue-Depth (RGB-D) data streams from Kinect V2 and constructs an end-to-end deep learning framework. Three-dimensional skeletal features are extracted via depthwise separable convolutions. A dynamic graph is built based on Graph Theory Matrix Analysis (GTMA); subsequently, spatiotemporal information is modeled by integrating graph convolution and temporal gated convolution. Memory prototype matching and collaborative attention mechanisms are introduced to generate personalized standards. These components facilitate action scoring, corrective feedback, and injury early warning. Furthermore, the study establishes a precise intelligent movement teaching and injury prevention system. The system models the spatial structural relationships and temporal variations of human movements to achieve intelligent analysis of motion postures and potential injury risks. In Study 1, the generalization ability of the system is verified utilizing the public NTU RGB[Formula: see text]D and NTU RGB[Formula: see text]D 120 datasets. The accuracy of cross-protocol recognition reaches 93.3%; meanwhile, movement scores are highly correlated with expert evaluations ([Formula: see text], [Formula: see text]). In Study 2, skeleton sequences are generated from public videos to verify the adaptability to real teaching scenarios. The intra-group variance of individual evaluations is reduced by approximately 60% compared with the fixed template method. Comprehensive experiments demonstrate that the system’s early warning window for typical injury-prone movements can reach 300 ms. Its early warning accuracy exceeds 93.5%, and the false alarm rate is below 3%, which is significantly related to subsequent abnormal postures ([Formula: see text] test, [Formula: see text]). SPOSE resolves the difficult problem of evaluating movement posture in high school physical education. Moreover, it reduces the risk of sports injuries and assists students in improving their athletic performance and safety. Thus, the intelligence and precision of physical education are promoted. The results demonstrate significant application potential for human kinematic analysis and motion injury risk identification.
科研通智能强力驱动
Strongly Powered by AbleSci AI