【作者】 王聪；

【导师】 张蕾； 裴一飞；

【作者基本信息】 天津大学 ， 计算机技术， 2019， 硕士

【摘要】 作为一种流行的人体生物识别技术,步态身份识别因其在安防监控和人机交互等领域中广阔的应用前景和经济价值,引起了研究人员的广泛关注。现有的步态身份识别技术大多依赖于专用的设备,由于受设备自身条件和使用环境的限制,无法全面满足不同应用场景的识别需求。利用Wi-Fi信号进行人体行为感知具有成本低、非入侵、普适性好等优势,可以用来弥补专用设备的不足。本文研究基于Wi-Fi信道状态信息(Channel State Information,CSI)的步态身份识别,主要解决了现有步态身份识别方法受行走路径限制的问题,并通过大量实验验证了本文实现系统的可靠性与稳定性。本文的研究内容和贡献主要包括两个方面:1.研究实现了基于Wi-Fi信道状态信息的步行动作识别。本文使用巴特沃斯带通滤波器来去除CSI中的高低频噪声和静态分量;为了消除数据的冗余、减少计算量,研究采用主成分分析对去噪后的CSI信号进行降维;根据功率谱密度在是否存在运动的情况下的不同表现,采用动态阈值的方法来判断人体运动的开始和结束;并从频域和时域两个角度提取完整的运动特征,来刻画不同的动作;为了实现稳定准确的步行动作识别,研究使用二分类的随机森林模型从日常众多的活动中检测步行动作;实验结果显示在多个复杂的室内环境中,本文实现了平均真正例率为96.68%、平均假正例率为4.61%的高检测精度。2.以步行动作识别为基础,研究实现了路径无关的步态身份识别。为了从与步行路径相关的CSI信号中获取路径无关的步态信息,本文研究根据CSI信号的测量特点,将身体的运动情况与测量信号的波动频率对应起来,并将其从正交的测量方向映射到行走路径上,使得无论受试对象沿着什么样的路径行走,都能将步态信息映射到对应路径上,进而获得路径无关的步态信息。为了提取充足有效的特征来刻画人体步态,研究集成使用手工特征提取方法和自动特征提取方法来保证特征的数量和质量。然后将所有特征输入到随机森林学习模型中,用来实现稳定、可靠、路径无关的步态身份识别;实验结果显示在识别人数达到50时,本文实现了精确的识别精度,Top-1精度为77.15%,Top-2精度为87.44%,Top-3精度为91.13%。更多还原

【Abstract】 As a popular biometric for human identification,gait identification has drawn extensive attention because of its broad application prospects and economic value in the fields of security monitoring and human computer interaction.Most of the existing gait identification technologies use specialized hardware.Due to the limitations of hardware’s conditions and application environment,they can not fully meet the requirements of different application scenarios.Wi-Fi based human behavior perception has the advantages of low cost,non-invasive and ubiquitous,which can be used to make up for the shortage of specialized hardware.In this paper,gait identification using Wi-Fi Channel State Information(CSI)is studied,which mainly solves the problem of walking path restriction in the existing gait identification method,and the reliability and stability of the system are verified by a large number of experiments.In summary,the main contributions are as follows:1.Based on the channel state information of Wi-Fi,walking activity recognition is studied.In this paper,Butterworth band-pass filter is used to remove the low and high frequency noise and static component in CSI;In order to eliminate the redundancy of data and reduce the computation cost,principal component analysis is used to reduce the dimension of the denoised CSI signal;According to the different performance of power spectral density under the condition of whether there is movement or not,dynamic threshold method is adopted to detect the beginning and end of human movement;And the complete movement features are extracted from time domain and frequency domain to describe different activities;In order to achieve a stable and accurate walking activity recognition,the random forest model with two-category is used to detect the walking activity from many daily activities;The experimental results show that in several complex indoor environments,a high accuracy of walking activity detection is achieved with mean true positive rate of 96.68%,and mean false positive rate of4.61%.2.Based on the walking activity recognition,the path independent gait identification is implemented.In order to obtain path independent gait information from CSI signal,according to the measurement characteristics of CSI signal,the relationship between the body movement and the fluctuation frequency of the measured signal is established.And,Mapping it from the orthogonal measurement directions to the walking path makes that no matter what shape of the walking path it is,the gait information can be mapped to the corresponding path to obtain the path independent gait information.In order to extract sufficient and effective features to describe the human gait,the manual feature extraction and automatic feature extraction are integrated to guarantee the quantity and quality of extracted features.Then,all the features are input into the random forest model to realize a stable and reliable path independent gait identification;The experiment results in typical indoor environment demonstrate the superior performance of the system,with Top-1,Top-2,and Top-3 identification accuracies of 77.15%,87.44%,and 91.13% respectively,when the number of subjects is 50.更多还原