基于智能感知的人体活动识别技术  

Human Activity Recognition Technology Based on Intelligent Perception

作　　者：周凯月 李佳[1] 乔树山[1] ZHOU Kaiyue;LI Jia;QIAO Shushan(Institute of Microelectronics of the Chinese Academy,Beijing 100029)

机构地区：[1]中国科学院微电子研究所,北京100029

出　　处：《智能感知工程》2024年第1期68-80,共13页

摘　　要：基于智能感知的人体活动识别(Human Activity Recognition,HAR)技术应用潜力巨大,尤其是在健康监测、智能运动和康复训练等领域。为了分析当前人体活动识别技术水平和未来发展方向,首先,阐述基于可穿戴传感器的智能传感技术;其次,归纳对比不同模态的公开数据集;再次,梳理智能算法研究现状,分析机器学习算法、深度学习算法和多模态算法在人体活动识别中的应用效果;最后,论述新型传感技术、感存算一体化架构及多模态方法等未来研究方向及主要挑战,如数据多样性、算法泛化能力和隐私保护等。Human activity recognition(HAR)technology based on intelligent perception has shown great application potential,especially in the fields of health monitoring,intelligent sports and rehabilitation training.In order to analyze the current level and future development direction of human activity recognition technology,firstly,the intelligent sensing technology based on wearable sensor is described.Secondly,the open data sets of different modes are summarized and compared.Thirdly,the research status of existing intelligent algorithms is summarized,and the application effects of machine learning algorithms,deep learning algorithms and multi-modal algorithms in human activity recognition are analyzed.Finally,the future research direction and main challenges of new sensing technology,integrated sensor-memory and computing architecture,and multi-modal approach are discussed,such as data diversity,algorithm generalization ability and privacy protection.

关 键 词：智能感知 可穿戴传感器 人体活动识别 深度学习 数据融合 健康监测 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP212[自动化与计算机技术—计算机科学与技术]