基于深度学习的跳绳姿态评估系统研究  

Research on jump rope action evaluation system based on deep learning

作　　者：丁瀚[1,2] 孙少明 孙怡宁[2] 张子康 彭伟[1,2] DING Han;SUN Shaoming;SUN Yining;ZHANG Zikang;PENG Wei(University of Science and Technology of China,Hefei 230026,China;Hefei Institutes of Physical Science,Chinese Academy of Sciences,Hefei 230031,China;Zhongke Hefei Technology Innovation Engineering Institute,Hefei 230088,China)

机构地区：[1]中国科学技术大学,安徽合肥230026 [2]中国科学院合肥物质科学研究院,安徽合肥230031 [3]中科合肥技术创新工程院,安徽合肥230088

出　　处：《电子设计工程》2025年第5期1-7,共7页Electronic Design Engineering

基　　金：国家重点研发计划(2018YFC2001304);安徽省重点研究与开发计划项目(202304a05020078);安徽省科技重大专项项目资助(202103a05020024)。

摘　　要：为解决青少年日常跳绳中可能存在的问题,规范其训练动作以降低受伤风险,提出了一种跳绳姿态检测与评估系统。该系统包含两个检测评估机制:机制1基于Attention-MLP神经网络模型进行标准化检测评估;机制2则对监测时段内高维跳绳特征数据进行提取与分析。系统包括数据获取、数据预处理以及跳绳姿态检测与评估。在数据预处理部分,文中进行了3种滤波算法的比较实验,结果表明,卡尔曼滤波对于跳绳的二维骨骼点数据平滑效果更佳。在标准化评估模块中,基于跳绳特征的数据集,搭建了Attention-MLP模型,并将其与常用的识别算法MLP、LSTM、CNN进行了实验对比。实验结果表明,Attention-MLP模型表现更为出色,准确率可达99.38%。In order to solve the possible problems in adolescents’daily rope skipping and to standardize their training movements to reduce the risk of injury,we propose a rope skipping posture detection and assessment system.The system consists of two detection and assessment mechanisms:mechanism 1 uses standardized detection and assessment based on the Attention-MLP neural network model,while mechanism 2 is based on the extraction and analysis of high-dimensional jumping rope features during the monitoring period.The components of the system include data acquisition,data preprocessing,and jump rope posture detection and evaluation.In the data preprocessing section,we conducted a comparative experiment of three filtering algorithms,and the results show that Kalman filtering is more effective for smoothing the 2D skeletal point data of jumping rope.In the standardized evaluation module,we built the Attention-MLP model based on the dataset of jumping rope features and conducted experiments to compare with the commonly used recognition algorithms MLP,LSTM and CNN.The experimental results show that the Attention-MLP model performs better,with an accuracy of 99.38%.

关 键 词：跳绳动作检测评估 骨骼点识别 数据滤波 深度学习 

分 类 号：TN-9[电子电信]