学习笔记(十五):基于WiFi的步态速度检测

19 篇文章 18 订阅

2020《GaitWay: Monitoring and Recognizing Gait Speed Through the Walls》读书笔记

  本文通过一对商用WiFi收发设备监测个人的步态速度,实现了自动识别稳定的步行周期提取物理上可信的和环境无关的速度特征,并据此识别受试者
  实验环境和结果:典型室内空间,对11名受试者进行了为期6个月的8个阶段的数据收集,结果是> 5000例步态实例,本文方法的速度估计误差中值为0.12 m/s, 90%为0.35 m/s,步长平均误差为3.36 cm。5人场景下身份识别准确率81.2%。
PS:个人感觉,散射模型利用了统计特性,牺牲速度的方向,获取速度大小的估计

1. 本文贡献

  1. 散射多径模型的基础上推出了一种用于速度估计的统计方法,该方法可以在目标超过10米远或在墙后时捕捉到准确的速度。
  2. 建立了步态速度监测和识别系统。通过提取一系列可信的物理特征,实现独立于位置、方向、环境和用户的服装来识别受试者。

2. 预备知识

步态周期(gait cycle)中的阶段划分:

在这里插入图片描述
步态特征:

  • 步态周期:每连续两步的平均持续时间 s
  • 步态速度(gait speed): m/s
  • 步幅(stride length): 来自速度的积分 m
  • 加速度:来自速度的倒数
  • 平滑度:谐波比(HR)量化给定步幅加速度的谐波组成来检查步幅内的步幅对称性。对每步的加速度进行离散傅里叶变换(DFT),HR为偶次谐波的振幅之和与奇次谐波的振幅之和之比。
  • 节奏性:应用滑动窗口来计算每个行走实例的一系列ACF,得到一个速度ACF矩阵。在此基础上,对每一列进行峰值检测,找出突出的峰值和相应的延迟。
  • 步态对称性:左右脚的步长均值和标准差作为特征。
  • 步态变异性: Recurrence Quantification Analysis (RQA) 当计算超过4个步态周期时,RQA达到一个稳定值

3. 速度估计

3.1 动态散射模型

在这里插入图片描述

  视人体为多散射体集合,向不同方向反射信号,并与被其他物体散射的信号通过多径叠加在Tx处。

  无线电信号被许多散射体散射,如墙壁、天花板、地板、家具、人体等。根据电磁波的叠加原理,CSI H(t, f)可分解为
H ( t , f ) = ∑ i ∈ Ω s ( t ) H i ( t , f ) + ∑ j ∈ Ω d ( t ) H j ( t , f ) + ε ( t , f ) H(t, f)=\sum_{i \in \Omega_{s}(t)} H_{i}(t, f)+\sum_{j \in \Omega_{d}(t)} H_{j}(t, f)+\varepsilon(t, f) H(t,f)=iΩs(t)Hi(t,f)+jΩd(t)Hj(t,f)+ε(t,f)CSI信号 = 静态散射Tx + 动态散射Tx + 噪声Tx(其方差为 σ 2 ( f ) \sigma^{2}(f) σ2(f)),且在足够短的时间内, 可 以 合 理 地 假 设 : \color{green}{可以合理地假设:} Ω s a n d Ω s \Omega_{s}and\Omega_{s} ΩsandΩs 近 似 不 变 \color{green}{近似不变} 。动态路径往往假定一个人体反射径为主导,动态散射能够利用统计特性。
H i ( t , f ) = ∫ 0 2 π F i ( θ , f ) exp ⁡ ( − j k v i cos ⁡ ( θ ) t ) d θ H_{i}(t, f)=\int_{0}^{2 \pi} F_{i}(\theta, f) \exp \left(-j k v_{i} \cos (\theta) t\right) d \theta Hi(t,f)=02πFi(θ,f)exp(jkvicos(θ)t)dθ即第i个散射体上的CSI贡献。 F i ( θ , f ) F_{i}(\theta, f) Fi(θ,f)为第 i i i个散射体从方向 θ θ θ上获得的MPC复增益, ∀ i ∀i i有如下特性:

  1. ∀ θ , F i ( θ , f ) ∀θ,F_{i}(\theta, f) θFi(θ,f)是具有相同方差 σ F i 2 ( f ) \sigma_{F_i}^{2}(f) σFi2(f) 圆对称高斯随机变量
  2. ∀ θ 1 ≠ θ 2 , F i ( θ 1 , f ) \forall \theta_{1} \neq \theta_{2}, F_{i}\left(\theta_{1}, f\right) θ1=θ2,Fi(θ1,f) and F i ( θ 2 , f ) F_{i}\left(\theta_{2}, f\right) Fi(θ2,f)are statistically
    independent;
  3. For ∀ i ≠ j ∈ Ω d , F i ( θ 1 , f ) \forall i \neq j \in \Omega_{d}, F_{i}\left(\theta_{1}, f\right) i=jΩd,Fi(θ1,f) and F j ( θ 2 , f ) F_{j}\left(\theta_{2}, f\right) Fj(θ2,f) are statistically independent for ∀ θ 1 \forall \theta_{1} θ1 and ∀ θ 2 \forall \theta_{2} θ2.

   H ( t , f ) H(t, f) H(t,f)的ACF记为 ρ H ( τ , f ) \rho_{H}(\tau, f) ρH(τ,f)(具体过程见论文):
ρ H ( τ , f ) = Cov ⁡ [ H ( t , f ) , H ( t + τ , f ) ] Cov ⁡ [ H ( t , f ) , H ( t , f ) ] = ∑ i ∈ Ω d σ F i 2 ( f ) J 0 ( k v i τ ) + σ 2 ( f ) δ ( τ ) ∑ i ∈ Ω d σ F i 2 ( f ) + σ 2 ( f ) \begin{aligned} \rho_{H}(\tau, f) &=\frac{\operatorname{Cov}[H(t, f), H(t+\tau, f)]}{\operatorname{Cov}[H(t, f), H(t, f)]} \\ &=\frac{\sum_{i \in \Omega_{d}} \sigma_{F_{i}}^{2}(f) J_{0}\left(k v_{i} \tau\right)+\sigma^{2}(f) \delta(\tau)}{\sum_{i \in \Omega_{d}} \sigma_{F_{i}}^{2}(f)+\sigma^{2}(f)} \end{aligned} ρH(τ,f)=Cov[H(t,f),H(t,f)]Cov[H(t,f),H(t+τ,f)]=iΩdσFi2(f)+σ2(f)iΩdσFi2(f)J0(kviτ)+σ2(f)δ(τ)其中 k = 2 π λ k=\frac{2 \pi}{\lambda} k=λ2π为波矢,意为2π长度上出现的全波数目,从相位的角度出发,可理解为:相位随距离的变化率(rad/m)。0阶贝塞尔函数 J 0 ( x ) = 1 2 π ∫ 0 2 π exp ⁡ ( − j x cos ⁡ ( θ ) ) d θ J_{0}(x)=\frac{1}{2 \pi} \int_{0}^{2 \pi} \exp (-j x \cos (\theta)) d \theta J0(x)=2π102πexp(jxcos(θ))dθ
  只有一个人在监控区域移动, 可 以 合 理 地 假 设 : \color{green}{可以合理地假设:} 所有由人引起的散射体的速度近似,i.e., v i = v v_{i}=v vi=v

躯干散射体(速度相近)贡献了大部分强散射信号,四肢影响小。

从而简化上式:
ρ H ( τ , f ) = ∑ i ∈ Ω d σ F i 2 ( f ) + σ 2 ( f ) δ ( τ ) ∑ i ∈ Ω d σ F i 2 ( f ) + σ 2 ( f ) J 0 ( k v τ ) ≜ α ( f ) J 0 ( k v τ ) \begin{aligned} \rho_{H}(\tau, f) &=\frac{\sum_{i \in \Omega_{d}} \sigma_{F_{i}}^{2}(f)+\sigma^{2}(f) \delta(\tau)}{\sum_{i \in \Omega_{d}} \sigma_{F_{i}}^{2}(f)+\sigma^{2}(f)} J_{0}(k v \tau) \\ & \triangleq \alpha(f) J_{0}(k v \tau) \end{aligned} ρH(τ,f)=iΩdσFi2(f)+σ2(f)iΩdσFi2(f)+σ2(f)δ(τ)J0(kvτ)α(f)J0(kvτ)其中 α ( f ) α(f) α(f)定义为每个子载波f的增益, J 0 ( k v τ ) J_{0}(k v \tau) J0(kvτ)定义为速度信号。该式直观表明了人体移动速度和CSI二阶统计量ACF之间的关系。
  实际中,CSI样本的ACF还需要引入一个估计噪声, ρ ^ H ( τ , f ) = α ( f ) J 0 ( k v τ ) + n ( τ , f ) \hat{\rho}_{H}(\tau, f)=\alpha(f) J_{0}(k v \tau)+n(\tau, f) ρ^H(τ,f)=α(f)J0(kvτ)+n(τ,f)

3.2 各个子载波上速度信号的最大比合并

S ( τ ) = ∑ f ∈ F w ⋆ ( f ) ρ ^ H ( τ , f ) S(\tau)=\sum_{f \in \mathcal{F}} w^{\star}(f) \hat{\rho}_{H}(\tau, f) S(τ)=fFw(f)ρ^H(τ,f)其中 S ( τ ) S(\tau) S(τ)为合并后的速度信号, w ⋆ ( f ) w^{\star}(f) w(f)为子载波频率为 f f f的子载波上的合并权重,其与 α ( f ) \alpha(f) α(f)成正比。

当适当组合所有子载波时,“好的”子载波将增强信号,而“坏的”子载波将有助于衰减噪声,因为它们的噪声项是独立的。

  利用 lim ⁡ τ → 0 J 0 ( k v τ ) = 1 \lim _{\tau \rightarrow 0} J_{0}(k v \tau)=1 limτ0J0(kvτ)=1 α ( f ) \alpha(f) α(f)约等于 ρ ^ H ( τ = 1 / F s , f ) \hat{\rho}_{H}\left(\tau=1 / F_{s}, f\right) ρ^H(τ=1/Fs,f)。所以 w ⋆ ( f ) w^{\star}(f) w(f)可直接取为 w ^ ⋆ ( f ) = ρ ^ H ( τ = 1 / F s , f ) \hat{w}^{\star}(f)=\hat{\rho}_{H}\left(\tau=1 / F_{s}, f\right) w^(f)=ρ^H(τ=1/Fs,f)

3.3 计算速度

在这里插入图片描述
  理论上,2(a)中组合速度信号的形状类似于贝塞尔函数 J 0 ( x ) ∣ x = k v τ J_{0}(x)_{|x=k v \tau} J0(x)x=kvτ,过匹配其关键特征,如第一个峰或谷的位置,可以提取出速度信号。以第一个峰值为例: v ^ = x 0 k τ ^ = x 0 λ 2 π τ ^ \hat{v}=\frac{x_{0}}{k \hat{\tau}}=\frac{x_{0} \lambda}{2 \pi \hat{\tau}} v^=kτ^x0=2πτ^x0λ其中 x 0 x_{0} x0其中为贝塞尔函数 J 0 ( x ) J_{0}(x) J0(x)第一个峰值对应的常数(由第0阶贝塞尔函数公式计算), τ ^ \hat{\tau} τ^为组合速度信号第一个峰值所对应的时滞。

  实际上,进一步采取两步提高速度估计。

  1. 为便于寻峰,使用组合速度信号的差值,如图2b所示,比图2d更为明显。在这种情况下, x 0 x_{0} x0成为 J 0 ( x ) J_{0}(x) J0(x)导数的第一个峰对应的位置。
  2. 利用两个接收天线之间的相位差来消除原始相位[40]的误差。

我不理解.jpg,2(c)和2(d)是怎么画出来的 ?
t t t时刻,时滞为 τ \tau τ时,组合速度信号的瞬时功率 ???

单子载波速度估计和MRC速度估计结果对比:
在这里插入图片描述

4. 步态特征提取和识别

4.1 识别稳定行走段

  步态分析和识别的第一步是确定一段稳定行走时期,在此期间受试者以习惯的步伐正常行走。
  当用户在平稳行走时,由于周期性的步伐节奏,观察到的速度会达到一定的重复模式范围。
  使用速度的ACF可测量这种行走周期。如图5所示,当用户稳定行走时,从速度的ACF可以看到明显的峰值。相比之下不同的步行速度时,ACF会更平坦。
  将滑动窗口(3秒)应用于速度估计,并计算每个窗口的ACF。然后采用峰值检测的ACF的速度,并检查第一个峰值。只有在观察到一系列连续可靠的峰值时,一个周期才被认为是稳定行走。
在这里插入图片描述
  为了增强鲁棒性,进一步引入步行速度的平均中心趋势。只有当平均速度大于某一数值(如0.7 m/s,小于1.0 m/s ~ 2.0 m/s的正常人类步行速度)时,才会使用步行周期进行步态分析。

4.2 步态周期提取

  对速度序列进行简单的峰值检测来识别step,为了对抗噪声和异常值,应用了某些限制条件(包括峰值突出度和高度)进行峰值检测。
  确定所有step后,通过删除第一个峰之前和最后一个峰之后的持续时间来修剪行走周期。

4.3 步态特征提取(略)

5. 实验设置

Tx-Rx 高度约1m,间隔8-11m,中间有墙
161 H40
视频获得groundtruth
文章中没有准确说明发射天线数量,同时接受天线数量至少为2

  • 2
    点赞
  • 14
    收藏
    觉得还不错? 一键收藏
  • 1
    评论
室内定位算法,有助于研究者认清研究方向,给定位算法一个准确的描述,是很好的参考教材。the possibility for the user to be notified by visible and Waveformof Distance audible warnings using buzzer and led ar to Very close to al 1903 Tme「 samples Fig 4 Distance measurement using the RSSi signal The system is used to determine the distance between Fig. 2. Tag4M Data Acquisition System tags and 2, 3, 4 or more APs. The tag scans after the aps ⅣV. EXPERIMENTS and sends the rssi values for each ap to the ap which is associated The system presented above can be used for objects or In order to read the corresponding rSsi values of all people localization in an indoor enviro found access points, a""operation is implemented at Let's suppose a building where several objects are Tag4M level. The time to scan is smaller than few distributed all over milliseconds, because the tag doesnt associate with the all For a better understanding of the system, it is showed an APs. These values are sent to a pC where the localization experiment where the objects are placed only at one floor. algorithm is implemented in LabVIFW20 10 (see Fig. 5) Each of these objects has attached a tag. The devices are Table I presents a scan operation result. In this case used to determine the location of the objects in the four APs(which are placed in the floor building)named building and the distance between tags and APs. To Hawk, Helicopter, Tag4M and WitagServer are detected measure the distance, the RSSI is processed. This method The corresponding rSSi values(in dBm) measured by the is not very accurate and is strong depended to the Tag4M are reported for every AP environment, but is very easy to implement with the TABLE 1. A SEQUENCE OF RESULTS FOR THE SCANOPERATION The Fig 3 presents the experiment environment for one EXFCUTED ON THE TAG4M DEVICE oor where the colored circles represents the APs and the rectangles represents the objects intended to be located This floor is divided in six rooms separated by walls RSSI Scan results: 4 Time Ssid Ch ad-Iloc Sec wps mac Address erp WMM SSI Supported Rates(Mbits, Mandatory) Crypto Suites CW Max/min AP SCAN. llawk. -57 dBm AP SCAN, Helicopter, -50 dBm, AP SCAN, Tag4M.-68 dBm AP SCAN. WitagServer. -45 dBm The Rssi values are converted to distance values using formula(1) The value of the received signal strength is a function of the transmitted power and the distance between the Fig 3. The Experiment Environment sender and the receiver. The received signal strength will decrease when the distance increases as the following First step in finding the location is distance equation shows [23] measurement using the RSSI signal. The experiment for computing the distance is presented in Fig. 4. The RSSI(IOn- logo+A)(1) experimental precisions of measurements are less than 5 meters Where: lal propagation constant al d propagation exponent. d represents the distance from sender, A represents the received signal strength at a distance of one meter worn even by people. The system runs on batterie w sSID having a characteristic life time of a couple of years and 回H offers a platform for sensor measurements. Thus, the system can use any existent infrastructure, with significant decrease of implementation costs REFERENCES [1] H. Liu, H. Darabi, P. Banerjee and J. Liu, ""Survey of wireless Distance 2 Tn] aRch Indoor Positioning Techniques and SystemS", IEEE TRANSACTIONS ON SYSTEMS. MAN. AND CYBERNETICS-PART C: APPLICATIONS AND REVIEWS VOL 37, NO 6, NOVEMBER 2007, pp. 1067-10801 Distance 3[] [2 Want, A. Hopper, V. FalcaO, J. Gibbons, "The Active Budge 4MDastDct v Location System, ACM Transactions on Information Systems vol. 10, no l, January 1992, pp 91-102. Fig. 5. Block Diagram of the localization program [3] Firefly Motion Tracking System User's guide", vcrsion 2.2 December 1999 http://www.gesturecentral.com/firefly/fireflyuserguide.pdf The application compute the distances between the Wi- [4] Northen Digilal Inc. Website, Oplotrak PROseries, 2011 Fi Tags placed on targets and the APs placed in indoor http:/www.ndigital.com/industrial/optotrakproseries-family.php [5] E. Aitenbichler, M. Mhlhuser, An /R Local Positioning System for environment. For this case, three APs are enough but the Smart Items and Devices Proc. 23rd IEEE Internationa number of Aps can be increased for obtaining more Conference on Distributed Computing Systems Workshops accurate position estimation. The distances between the TWSAWC03), 2003 Wi-fiTagsandeachAparerepresentedascirclesasit[6]topaz,2004,http:/www.tadlys.co.il/pAges/productcontent.asp?igi bald=2 can be seen in Fig. 6, having the centre in the AP locations [7 A. Kotanen, M. Hannikainen, H. Leppakoski, and T. D which have previously known coordinates. The target's Hamalaincn, E.cperiments on local positioning with Bluetoothin position is given by the point at the intersection of the Proc. IEEE Int Conf. Inf. Technol. Comput. Commun., Apr 2003 three circles PP.297-303 [8 J. Hallberg, M. Nilsson, and k. Synnes, "Positioning with Bluetooth"in Proc. Ieee 10th Int Conf. Telecommun. Mar. 2003 ol.2,pp.954958 9 Active bat website. 2008 http:/www.cl.cam.ac.uk/research!dtg/attarchive/bat/ [10] Hazas, M, Hopper, A, "A Novel Brucdbund Ultrasonic Locution System for Improved Indoor Positioning IEEE Transactions on mobile Computing. Vol 5, No 5, May 2006 [11 N B Priyantha, The Cricket Indoor Location System", PhD thes MIT.2005 2]SonitorSystemWebsite2011,http://www.sonitor.com/ [13] Minami M, Fukuju Y, Hirasawa K, Yokoyama S, Mizumachi M Aoyama T ,"Dolphin. A practical approach for implementing a fully distributed indoor ultrasonic positioning system", Ubicomp, 2004, 347-365 Fig. 6. Front Panel of the localization program 14]J. Ilightower, R. Want, and G. borriello, SpotON: An indoor 3D cation sensing technology hased on RF signal strength"Univ The simplicity and cost efficiency of the rssi based Washington, Seattle, Tech Rep UW CSE 2000-02-02, Feb. 2000 localization make the proposed system a desired candidate 15]L. M. Ni, Y. Liu, Y. C. Lau, and A. P. Patil,"LANDMARC: Indoor location sensing using active RFID Wireless Netw., vol 10, no 6 for specific applications like tracking and positioning svstems hP.701-70.N0v.200 Zebra Technologies Corporation eb sit http://www.7ebra.com/id/zebra/na/en/index/products/location/isoie c 24730 2html V. CONCLUSION [17 P. Bahl and V. Padmanabhan, "RADAR: An in-building rf based user location and tracking system, Proc. IEEE INFOCOM, voL 2 In this paper, an indoor localization system based on March200,pp.775784 ultra-low power Wi-Fi technology and designed to 18Ekahau2011,http://www.ekahau.com: [19] T. King, S. Kopf, T. Haenselmann, C. Lubberger and w determine location of objects in a closed environment is Effelsberg, "COMPASS: A Probabilistic Indoor Positioning System proposed gILo rOC In the first part of this paper a survey of existing indoor Workshop on Wireless Network Testbeds, Experimental evaluation and Cllaracterization (WiNTECID), Los Angeles, CA, USA positioning systems was made In the second part was presented the way in which an [20] Convert sensor data to web pages using a Cloud Instrument, june object can be localized by using access points and Tag 4M 2011.http://www.tag4m.com devices which may read the rssi values [21Tag4mDatasheet2010,http://test.tag4m.com/wp- content/uploads/20/03/Tag4M Prod Datasheet Revised3. pdf The importance of the proposed system lies in ultra-low [22] Aamodt, K.(2006). CC2431 Location Engine, Application Note power and Wi-Fi transmission capabilities that are An042,fromhttp://focus.tii.co.ip/ip/lit/an/swra095/swra095.pdf embedded in a very small package, why they are suitable 320

“相关推荐”对你有帮助么?

  • 非常没帮助
  • 没帮助
  • 一般
  • 有帮助
  • 非常有帮助
提交
评论 1
添加红包

请填写红包祝福语或标题

红包个数最小为10个

红包金额最低5元

当前余额3.43前往充值 >
需支付:10.00
成就一亿技术人!
领取后你会自动成为博主和红包主的粉丝 规则
hope_wisdom
发出的红包
实付
使用余额支付
点击重新获取
扫码支付
钱包余额 0

抵扣说明:

1.余额是钱包充值的虚拟货币,按照1:1的比例进行支付金额的抵扣。
2.余额无法直接购买下载,可以购买VIP、付费专栏及课程。

余额充值