基于移动设备运动传感器的人体行为识别算法研究
发布时间：2018-12-07 17:14
【摘要】：可穿戴计算以一种“人机交互”的新型计算模式,实现了人类社会与信息环境、物理环境的连通,在军事、公共卫生、电子消费、体育、教育等众多领域都有着广泛的应用,现已成为学术界和工业界的研究热点。人体行为识别技术是可穿戴计算中重要的研究分支。本文针对基于运动传感器的人体行为识别算法开展创新性工作,对加速度传感器的数据采集、数据预处理,机器学习分类模型以及行为识别算法展开研究,主要研究内容包括:(1)探索了利用智能手机内置的运动传感器采集人体行为活动数据,对获得的三轴加速度传感器数据进行预处理,去除噪声并对数据进行片段分割;(2)针对人体行为活动中典型的行走、上楼、下楼、坐、站立和跌倒这六种行为,分别进行了时域和频域特征提取,并对这六个行为的时域和频域特征两两对比分析,得到了更深层次的特征区分细节,建立了各人体行为活动的特征数据集;(3)在传统人体行为识别支持向量机模型的基础上,将支持向量分类机与二次核函数理论相结合,通过理论分析构造出了二次核支持向量分类机模型,进而提出了基于二次核支持向量分类机模型的人体行为识别算法,与现有的基于随机森林分类模型的人体行为识别算法相比,本文算法的识别精度更高。
[Abstract]:Wearable computing has been widely used in many fields, such as military affairs, public health, electronic consumption, sports, education and so on, because of its new computing mode of "human-computer interaction", which realizes the connection between human society and information environment, physical environment and physical environment. It has become a research hotspot in academia and industry. Human behavior recognition is an important branch of wearable computing. This paper focuses on the innovative work of human behavior recognition algorithm based on motion sensor, and researches on acceleration sensor data acquisition, data preprocessing, machine learning classification model and behavior recognition algorithm. The main research contents are as follows: (1) using the motion sensor built in the smart phone to collect human behavior data, preprocessing the obtained three-axis acceleration sensor data, removing noise and segmenting the data; (2) aiming at the typical walking, going upstairs, going downstairs, sitting, standing and falling, the feature extraction in time domain and frequency domain is carried out respectively, and the characteristics of time domain and frequency domain of these six behaviors are compared and analyzed. The deeper feature distinguishing details are obtained, and the feature data sets of human behavior are established. (3) based on the traditional support vector machine model of human behavior recognition, a quadratic kernel support vector classifier model is constructed by combining the support vector classification machine with the quadratic kernel function theory. Furthermore, a human behavior recognition algorithm based on quadratic kernel support vector classifier model is proposed. Compared with the existing human behavior recognition algorithm based on stochastic forest classification model, the recognition accuracy of this algorithm is higher.
【学位授予单位】：宁夏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP212
[Abstract]:Wearable computing has been widely used in many fields, such as military affairs, public health, electronic consumption, sports, education and so on, because of its new computing mode of "human-computer interaction", which realizes the connection between human society and information environment, physical environment and physical environment. It has become a research hotspot in academia and industry. Human behavior recognition is an important branch of wearable computing. This paper focuses on the innovative work of human behavior recognition algorithm based on motion sensor, and researches on acceleration sensor data acquisition, data preprocessing, machine learning classification model and behavior recognition algorithm. The main research contents are as follows: (1) using the motion sensor built in the smart phone to collect human behavior data, preprocessing the obtained three-axis acceleration sensor data, removing noise and segmenting the data; (2) aiming at the typical walking, going upstairs, going downstairs, sitting, standing and falling, the feature extraction in time domain and frequency domain is carried out respectively, and the characteristics of time domain and frequency domain of these six behaviors are compared and analyzed. The deeper feature distinguishing details are obtained, and the feature data sets of human behavior are established. (3) based on the traditional support vector machine model of human behavior recognition, a quadratic kernel support vector classifier model is constructed by combining the support vector classification machine with the quadratic kernel function theory. Furthermore, a human behavior recognition algorithm based on quadratic kernel support vector classifier model is proposed. Compared with the existing human behavior recognition algorithm based on stochastic forest classification model, the recognition accuracy of this algorithm is higher.
【学位授予单位】：宁夏大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP212
【参考文献】
相关期刊论文 前10条
1 吴婷;周宇;;基于BP神经网络的人体行为识别方法研究与实现[J];教育教学论坛;2016年51期
2 王赛迪;;基于人工神经网络的人体识别算法研究[J];电子世界;2016年17期
3 苟涛;;基于可穿戴式三轴加速度传感器的人体行为识别[J];自动化应用;2015年12期
4 刘欣然;;老人跌倒监测定位装置的研究[J];无线电通信技术;2016年01期
5 卢先领;徐仙;;基于加速度与HGA-BP神经网络的人体行为识别[J];计算机工程;2015年09期
6 苏z延，
本文编号：2367536
本文链接：https://www.wllwen.com/kejilunwen/zidonghuakongzhilunwen/2367536.html
