无线体域网中人体动作监测与识别若干方法研究
发布时间：2018-12-13 03:09
【摘要】：无线体域网是由可感知人体多种生理参数的轻便、可穿戴或可植入的传感器节点构建的无线网络。无线体域网为人体健康监测提供了新的手段,在疾病监控、健康恢复、特殊人群监护等领域有着巨大的应用意义和需求。通过佩戴在身体上的微惯性传感器,体域网可以采集人体的运动信号,在人体动作监测方面得到广泛应用,可实现人体动作识别、异常动作检测、步态识别与分析、运动能耗分析等目的。 在利用无线体域网进行人体运动监测过程中,如何在满足身体活动监测指标要求的同时提高传感器节点的能量有效性,以便能在实际应用中长时间不间断地进行人体动作监测,是一个具有挑战性的问题。本文以由多个可穿戴的微惯性传感器构成的无线体感网为研究对象,围绕能量有效性,以稀疏表示和压缩感知理论为主线,从信号识别、信号压缩、数据融合、功率控制这四个方面展开研究。主要工作和创新点如下： (1)提出了一种基于自学习稀疏表示的动态手势识别方法L-SRC.针对手势识别中手势长短不一的问题,将手势样本向量进行归一化线性插值,从而将手势识别问题转化为求解待识别样本在训练样本中的稀疏表示问题；针对如何提高手势识别精度和速度的问题,采用基于类别的字典学习方法寻求一个较小的并经过优化的超完备字典来计算待识别样本的稀疏表示,从而在手势识别阶段大幅度缩减识别算法的计算复杂度,满足快速识别要求。在包含18种手势的数据集上验证了提出的L-SRC手势识别方法在保证识别精度的同时提升了识别速度。 (2)提出了两种压缩分类的动作识别方法RP-CCall和RP-CCeach.针对运动信号的时间冗余性和稀疏性,结合压缩感知和稀疏表示理论,将传感信号压缩与动作识别相结合,以满足一定动作识别率的同时降低传感器节点的能耗。两种RP-CC方法是在传感器节点上利用随机投影对运动信号进行数字化的压缩采样,通过减少无线体域网的数据传输量来节省能耗；在基站上直接对压缩的数据建立稀疏表示的人体运动模式识别模型,利用稀疏系数的分布来实现动作识别。理论分析了压缩分类动作识别方法能正确识别的基本条件。找到了能在存储和计算资源有限的传感器节点上实现的随机投影矩阵。在包含13种动作的数据集上进行了验证,结果显示RP-CCall方法和RP-CCeach方法在对压缩的数据识别时也能达到无压缩时相近似的识别准确率,并高于最近邻、支持向量机等分类方法。 (3)提出了基于分布式压缩感知和联合稀疏表示的动作识别方法DCS-JSRC.针对无线体域网中多传感器采集的运动数据之间的时空相关性,采用分布式压缩感知在传感器节点进行分布式压缩,充分利用这种相关性来进一步压缩数据以降低传输能耗。在基站通过探索多传感器节点感知运动信号的时空相关性,构建适用于动作识别的联合稀疏表示模型,将多传感器的动作识别问题转化为多变量稀疏线性回归问题来求解。采用层次贝叶斯模型来求解稀疏表示系数,利用不同传感器节点的相互关联来进一步提高动作识别的准确率。在动作数据集上进行验证,实验结果显示DCS-JSRC方法在相同压缩比的情况下获得了比RP-CCall方法和RP-CCeach方法更高的识别准确率。 (4)设计了轻量级的基于动作行为的自适应功率反馈控制机制PID-A。针对无线体域网中链路通信质量受人的运动、姿态变化影响具有动态时变特性,通过实测人体不同动作以及发射功率变化对无线链路的影响,分析和总结了在人体不同运动状态下节点的发射功率与链路通信质量的变化特性和规律,在此基础上建立基于反馈的功率控制系统模型,结合人体动作识别的结果,来动态调整无线体域网中节点的发射功率。实验结果显示PID-A功率控制机制可保证在数据包成功接收的条件下降低了传感器节点发送数据包的平均能耗。 (5)为了验证算法在实际系统中的性能,设计并实现了用于人体运动监测的无线体域网原型系统。利用所构建的基于微惯性传感器的无线体域网,采集人体在日常活动中的动作信号,实际验证了所提出的动作识别算法的识别准确率,并对传感器节点的能耗进行了分析,验证了算法的能量有效性。
[Abstract]:the wireless body domain network is a wireless network constructed from a light, wearable or implantable sensor node that can sense a variety of physiological parameters of the human body. The wireless body area network provides new means for human health monitoring, and has great application meaning and requirement in the fields of disease monitoring, health recovery, special crowd monitoring and the like. By wearing the micro-inertial sensor on the body, the body-domain network can collect the motion signal of the human body, and has wide application in human motion monitoring, and can realize the purposes of human body motion identification, abnormal motion detection, gait recognition and analysis, motion energy consumption analysis, and the like. In the process of human motion monitoring by using the wireless body domain network, how to improve the energy efficiency of the sensor nodes while meeting the requirements of the physical activity monitoring indexes, so as to be able to carry out human motion monitoring for a long time in the practical application, is a challenging question, In this paper, a wireless body-sensing network composed of a plurality of wearable micro-inertial sensors is used as the research object, and the energy efficiency is focused on the basis of the sparse representation and the compression-sensing theory, and the four aspects of signal identification, signal compression, data fusion and power control are developed. Research. Key work and innovative points such as (1) A dynamic gesture recognition method based on self-learning sparse representation is proposed. SRC. The gesture recognition problem is transformed into a sparse representation problem for solving the sample to be identified in the training sample, and the gesture recognition problem is converted into a sparse representation problem in the training sample for the sample to be identified; and the gesture recognition precision and the speed are improved. According to the problem, the sparse representation of the sample to be identified is calculated by adopting a category-based dictionary learning method, so that the calculation complexity of the identification algorithm is greatly reduced in the gesture recognition stage, and the rapid recognition is met. The proposed L-SRC gesture recognition method is verified to improve the recognition precision while the recognition accuracy is guaranteed. (2) Two types of motion recognition methods, RP-CCall, and RP-C, are proposed. Cach. Combining the time redundancy and sparsity of the motion signal, combining the compression-aware and sparse representation theory, the sensing signal compression is combined with the action recognition to meet the recognition rate of a certain action while reducing the sensor. The method comprises the following steps of: carrying out digital compression sampling on a motion signal on a sensor node by using a random projection on a sensor node, saving energy consumption by reducing the data transmission amount of the wireless body domain network, a pattern recognition model that uses the distribution of the sparse coefficients to The recognition of the present action is carried out. The theoretical analysis of the identification of the motion recognition method of the compression classification can be correctly identified. The basic condition of a sensor node that can be realized on a sensor node with limited storage and computational resources The results show that the RP-CCall method and the RP-CCeach method can achieve similar recognition accuracy when the compressed data is not compressed, and it is higher than the nearest neighbor and support vector machine. (3) An action identification method based on distributed compression-aware and combined sparse representation is presented. CS-JSRC. The spatial and temporal correlation between the motion data collected by multiple sensors in the wireless body domain network is distributed in the sensor node by the distributed compression sensing, and the correlation is fully utilized to further compress the data. in that base station, the time-space correlation of the motion signal is sensed by the base station, a joint sparse representation model suitable for action identification is constructed, and the action identification problem of the multi-sensor is converted into a multi-variable sparse linear model, The problem of regression is solved. A hierarchical Bayesian model is used to solve the sparse representation coefficient, and the correlation of different sensor nodes is used to further improve the motion. The results show that the method of the DCS-JSRC is more effective than the RP-CCall method and the RP-CCeach method in the case of the same compression ratio. High recognition accuracy. (4) A lightweight self-adaptive power feedback based on action behavior is designed The control mechanism PID-A. Aiming at the movement of the link communication quality in the wireless body domain network, the influence of the attitude change has the dynamic time-varying characteristic, and by actually measuring the different actions of the human body and the transmission power change, The influence of the wireless link on the wireless link is analyzed and summarized, the change characteristics and the law of the transmission power and the link communication quality of the nodes in different motion states of the human body are analyzed and summarized, a power control system model based on the feedback is established, the results of human motion recognition are used to dynamically adjust the wireless volume domain, The experimental results show that the PID-A power control mechanism can ensure that the sensor node is reduced under the condition that the data packet is successfully received. The average energy consumption of the data packet is sent. (5) In order to verify the performance of the algorithm in the real system, it is designed and implemented for human motion monitoring. based on the built wireless body domain network of the micro-inertial sensor, the motion signal of the human body in the day-to-day activity is collected, the identification accuracy of the proposed action identification algorithm is actually verified, and the energy consumption of the sensor node is analyzed,
【学位授予单位】：湖南大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TN92;TP391.41
本文编号：2375766
[Abstract]:the wireless body domain network is a wireless network constructed from a light, wearable or implantable sensor node that can sense a variety of physiological parameters of the human body. The wireless body area network provides new means for human health monitoring, and has great application meaning and requirement in the fields of disease monitoring, health recovery, special crowd monitoring and the like. By wearing the micro-inertial sensor on the body, the body-domain network can collect the motion signal of the human body, and has wide application in human motion monitoring, and can realize the purposes of human body motion identification, abnormal motion detection, gait recognition and analysis, motion energy consumption analysis, and the like. In the process of human motion monitoring by using the wireless body domain network, how to improve the energy efficiency of the sensor nodes while meeting the requirements of the physical activity monitoring indexes, so as to be able to carry out human motion monitoring for a long time in the practical application, is a challenging question, In this paper, a wireless body-sensing network composed of a plurality of wearable micro-inertial sensors is used as the research object, and the energy efficiency is focused on the basis of the sparse representation and the compression-sensing theory, and the four aspects of signal identification, signal compression, data fusion and power control are developed. Research. Key work and innovative points such as (1) A dynamic gesture recognition method based on self-learning sparse representation is proposed. SRC. The gesture recognition problem is transformed into a sparse representation problem for solving the sample to be identified in the training sample, and the gesture recognition problem is converted into a sparse representation problem in the training sample for the sample to be identified; and the gesture recognition precision and the speed are improved. According to the problem, the sparse representation of the sample to be identified is calculated by adopting a category-based dictionary learning method, so that the calculation complexity of the identification algorithm is greatly reduced in the gesture recognition stage, and the rapid recognition is met. The proposed L-SRC gesture recognition method is verified to improve the recognition precision while the recognition accuracy is guaranteed. (2) Two types of motion recognition methods, RP-CCall, and RP-C, are proposed. Cach. Combining the time redundancy and sparsity of the motion signal, combining the compression-aware and sparse representation theory, the sensing signal compression is combined with the action recognition to meet the recognition rate of a certain action while reducing the sensor. The method comprises the following steps of: carrying out digital compression sampling on a motion signal on a sensor node by using a random projection on a sensor node, saving energy consumption by reducing the data transmission amount of the wireless body domain network, a pattern recognition model that uses the distribution of the sparse coefficients to The recognition of the present action is carried out. The theoretical analysis of the identification of the motion recognition method of the compression classification can be correctly identified. The basic condition of a sensor node that can be realized on a sensor node with limited storage and computational resources The results show that the RP-CCall method and the RP-CCeach method can achieve similar recognition accuracy when the compressed data is not compressed, and it is higher than the nearest neighbor and support vector machine. (3) An action identification method based on distributed compression-aware and combined sparse representation is presented. CS-JSRC. The spatial and temporal correlation between the motion data collected by multiple sensors in the wireless body domain network is distributed in the sensor node by the distributed compression sensing, and the correlation is fully utilized to further compress the data. in that base station, the time-space correlation of the motion signal is sensed by the base station, a joint sparse representation model suitable for action identification is constructed, and the action identification problem of the multi-sensor is converted into a multi-variable sparse linear model, The problem of regression is solved. A hierarchical Bayesian model is used to solve the sparse representation coefficient, and the correlation of different sensor nodes is used to further improve the motion. The results show that the method of the DCS-JSRC is more effective than the RP-CCall method and the RP-CCeach method in the case of the same compression ratio. High recognition accuracy. (4) A lightweight self-adaptive power feedback based on action behavior is designed The control mechanism PID-A. Aiming at the movement of the link communication quality in the wireless body domain network, the influence of the attitude change has the dynamic time-varying characteristic, and by actually measuring the different actions of the human body and the transmission power change, The influence of the wireless link on the wireless link is analyzed and summarized, the change characteristics and the law of the transmission power and the link communication quality of the nodes in different motion states of the human body are analyzed and summarized, a power control system model based on the feedback is established, the results of human motion recognition are used to dynamically adjust the wireless volume domain, The experimental results show that the PID-A power control mechanism can ensure that the sensor node is reduced under the condition that the data packet is successfully received. The average energy consumption of the data packet is sent. (5) In order to verify the performance of the algorithm in the real system, it is designed and implemented for human motion monitoring. based on the built wireless body domain network of the micro-inertial sensor, the motion signal of the human body in the day-to-day activity is collected, the identification accuracy of the proposed action identification algorithm is actually verified, and the energy consumption of the sensor node is analyzed,
【学位授予单位】：湖南大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TN92;TP391.41
【参考文献】
相关期刊论文 前9条
1 王万良;杨经纬;蒋一波;;基于运动传感器的手势识别[J];传感技术学报;2011年12期
2 张志强;黄志蓓;吴健康;;髋关节角多模型贝叶斯动态估计[J];电子与信息学报;2011年04期
3 方红;章权兵;韦穗;;基于亚高斯随机投影的图像重建方法[J];计算机研究与发展;2008年08期
4 宫继兵;王睿;崔莉;;体域网BSN的研究进展及面临的挑战[J];计算机研究与发展;2010年05期
5 梁秀波;张顺;李启雷;张翔;耿卫东;;运动传感驱动的3D直观手势交互[J];计算机辅助设计与图形学学报;2010年03期
6 胡正平;宋淑芬;;基于类别相关近邻子空间的最大似然稀疏表示鲁棒图像识别算法[J];自动化学报;2012年09期
7 李方敏;徐文君;刘新华;;无线传感器网络功率控制技术[J];软件学报;2008年03期
8 陆希玉;陈鑫磊;孙光;金德鹏;苏厉;曾烈光;;超宽带体域网信道测量及传输特性分析[J];清华大学学报(自然科学版);2011年11期
9 孙洪;张智林;余磊;;从稀疏到结构化稀疏:贝叶斯方法[J];信号处理;2012年06期
相关博士学位论文 前3条
1 薛洋;基于单个加速度传感器的人体运动模式识别[D];华南理工大学;2011年
2 刘蓉;人体运动信息获取及物理活动识别研究[D];华中科技大学;2009年
3 姜鸣;基于体感网的人体动作监测识别的研究[D];大连理工大学;2012年
，本文编号：2375766
本文链接：https://www.wllwen.com/kejilunwen/wltx/2375766.html
