Feng Lin, Chen Song, Yan Zhuang, Wenyao Xu, Changzhi Li, Kui Ren

Presentation on theme: "Feng Lin, Chen Song, Yan Zhuang, Wenyao Xu, Changzhi Li, Kui Ren"— Presentation transcript:

1 Cardiac Scan: A Non-contact and Continuous Heart-based User Authentication System
Feng Lin, Chen Song, Yan Zhuang, Wenyao Xu, Changzhi Li, Kui Ren CSE, University at Buffalo, SUNY

2 Motivation cardiac motion Continuous authentication 指纹，密码：需要用户主动认证
人脸（凝视，虹膜），敲击键盘：对用户有一定的动作要求 cardiac motion

3 Heart-based User Authentication System

4 Design Goals Intrinsic liveness Unobtrusive Authentication
Highly Secure Cost-effective and Easy-to-Use Resilient to Background Noise and Use Conditions 内在活性：需要鉴别属性是否来自于真实的用户，这可以预防重放攻击 不引人注目：用户正常状态下就可以进行验证，无须主动配合 高安全性：具有唯一性，很难被伪造或者盗取 成本有效和易用：一些生物特征的获取需要昂贵的专用设备，Cardiac Scan使用的是雷达感应器 抗干扰能力强

5 Non-contact Cardiac Motion Sensing
Rationale 具有基础理论支持 五个阶段： 1、心室充血 2、心房收缩 3、定容心室收缩 4、心室注射 5、定容心室舒张 这五个阶段，心脏的3D形变具有显著的不同，而且每个人具有独一无二性

6 Non-contact Cardiac Motion Sensing
Feasibility Respiration and heartbeat rates Existing cancellation approaches are not applicable DC-coupled continuous-wave (CW) Doppler radar sensors 没有消除身体运动 现有的方法会牺牲信号的质量，不能在实际中使用，获得心脏的运动

7 Architecture with Dynamic DC Tuning
Beam control Smart DC tuning 利用beamforming，控制发射天线的赋值和相位，使得到达的信号比较强 shifter and attenuator add a portion of the transmitted signal to the receiver signal A baseband fine-tuning block was implemented to dynamically adjust the amplifier bias 2.4-40GHz Optimal carrier frequency

8 System Overview Pre-processing one-pass noise reduction techniques
adaptive noise canceling techniques Butterworth bandpass filter LMS自适应滤波器

9 De-noising-aware Radar Demodulation
发送和接收信号 I路和Q路信号

10 Heart Geometric Features
Fiducial descriptors AFP：第一个最大点 VFP：第二个最大点 ASP：AFP和VFP之间的最小值

11 Heart Geometric Features
Fiducial-based descriptors 基准点的抽取对于刻画心脏运动的特征特别重要 由于各种噪音的存在，曲率半径更加鲁棒

12 Continuous Authentication Protocol
Continuous Authentication Parameters Continuous Authentication Scenarios Refreshing interval False negative tolerance threshold

13 Experimental Setting Doppler radar has two antennas with the beam width of 45 degrees 2.4 GHz with the bandwidth of 5 kHz 78 healthy subjects, 20 trials for 8s A subject is sitting one meter away from both radar sensors, a chest belt and a pulsed sensor is attached to the subject. DTW-相似性度量， SVM-分类， 十折交叉验证 random body motion creates an opposite Doppler frequency shift to the signals of the radars, while cardiac motion leads to the same polarity

14 Accuracy Balanced Accuracy and F-measure
精确率表示预测为正的样本中有多少是真正的正样本，召回率表示的是样本中的正例有多少被预测正确。

15 Authentication Time Authentication with 4 sec duration is the optimal choice

16 Evaluation in Complex Conditions

17 Vulnerability Study 1、the direction of arrival (DoA) 2、vital sign
呼吸和身体摆动 12 replay attacks were rejected by our liveness detection method

18 Summary 1、We explore new cardiac motion-based biometrics for continuous authentication. The approach holds the potential to transform existing authentication systems into a more undeceivable, disclosure-resistant and user-friendly solution. 2、We evaluate Cardiac Scan through a set of multi-scene evaluations. 难以伪装、防止泄露、用户友好

19 思考 更偏底层，更扎实（心脏部分的介绍、硬件修改） 实验很充分（78个用户、考虑了很多影响因素） 写作技巧很重要
文章比较偏安全，跟我们的研究思路不太一样 系统设计介绍的不够充分（数据太少） 利用超声波做用户验证？

20 Thanks!