健身行为高精度监测和个性化精准健身方案智能推送技术  

Research on High Precision Monitoring of Fitness Behavior and Personalized Precision Fitness Program Intelligent Push Technology

作　　者：张晋喜 李振 李沫含 邱华龙 刘程林 杨铭 徐鲁友 贾壮 周志雄[1] ZHANG Jinxi;LI Zhen;LI Mohan;QIU Hualong;LIU Chenglin;YANG Ming;XU Luyou;JIA Zhuang;ZHOU Zhixiong(Capital University of Physical Education and Sports,Beijing 100191,China;Fujian Normal University,Fuzhou,Fujian 350070,China)

机构地区：[1]首都体育学院,北京100191 [2]福建师范大学,福建福州350070

出　　处：《首都体育学院学报》2023年第3期276-283,共8页Journal of Capital University of Physical Education and Sports

基　　金：国家重点研发计划项目(2020YFC2006200)。

摘　　要：世界卫生组织建议成年人每周进行150~300 min中等强度或75~150 min高强度身体活动,身体活动量化监测和个性化健身方案是科学健身的重要基础。针对健身行为高精度监测和个体运动方案智能推送缺乏有效的方法,建立基于卷积神经网络和长短时记忆网络模型的身体活动强度预测算法模型,对三轴加速度计采集的运动数据序列进行特征提取和时序相关性分析,预测值与真实值之间的平均绝对百分比误差为12.03%,均方误差为1.027;研制基于用户的身体活动、运动风险、健康水平、健身目标等特征指标的智能生成个性化运动方案算法,基于“首体健身”健身指导系统将运动方案与标签化的动作库和健身课程视频按照关联规则的匹配方法组成可视化健身方案。基于卷积神经网络和长短时记忆网络的算法模型预测身体活动强度的精度高,基于身体活动量、运动风险等特征智能推送的个性化健身方案可满足居民健身需求。The World Health Organization recommends that adults engage in 150~300 minutes of moderate-intensity or 75~150 minutes of vigorous-intensity physical activity per week.Quantitative monitoring of physical activity and personalized fitness prescriptions are important foundations of scientific fitness.There is a lack of effective methods for continuous high-precision monitoring of fitness behavior and intelligent delivery of individual exercise prescription.This study establishes a physical activity intensity prediction algorithm model based on convolutional neural networks and long short-term memory networks.It extracts features from motion data sequences collected by triaxial accelerometers and analyzes their temporal correlations.The average absolute percentage error between predicted values and actual values is 12.03%,and the mean squared error is 1.027.An algorithm is developed to generate personalized exercise prescriptions intelligently based on user characteristics such as physical activity,exercise risk,health level,and fitness goals.The visual fitness prescription is created by matching the exercise prescriptions with the tagged motion library and fitness course videos using the association rule in the‘SHOUTI Fitness’exercise guidance system.The accuracy of predicting physical activity intensity using the algorithm model based on convolutional neural networks and long short-term memory networks is high.The personalized fitness prescription,based on features such as physical activity level and exercise risk,intelligently delivered to users,meets the needs of the general public for fitness.

关 键 词：身体活动 代谢当量 运动处方 神经网络 

分 类 号：TP3[自动化与计算机技术—计算机科学与技术]