210 likes | 222 Views
Explore WiTalk, a CSI-based motion sensing scheme for fine-grained motion detection using WiFi. Discover how WiTalk enables context-free motion sensing and its applications in various scenarios.
E N D
Context-Free Fine-Grained Motion Sensing using WiFi Changlai Du, Xiaoqun Yuan, Wenjing Lou, Thomas Hou Virginia Tech Wuhan University June. 13, 2018
Ubiquitous Wireless • Wireless communications are ubiquitous • More wireless devices connect Internet • 20 billion will be in use by 2020[1] • 99% connect to Internet[2] [1] Gartner, http://www.gartner.com/newsroom/id/3598917 [2] The Future of Wireless. http://circuitcellar.com/cc-blog/iot-the-future-of-wireless-connect-anywhere-solutions/
Transferring Data 010110101011 110001010101
WiTalk: Fine-Grained Motion Sensing • CSI-based Fine-grained motion sensing Scheme • One-time training, resilient to context change • Work accepted to SECON 18 • C. Du, X. Yuan, W. Lou, Y.T. Hou, Context-Free Fine-Grained Motion Sensing using WiFi. IEEE SECON 2018, accepted.
Literature Review • CSI based motion sensing has been used different fields • Human localization: ACM Computing Surveys13 • Activity detection: MobiCom15, MobiCom14 • Human authentication: Ubicomp16, IPSN16 • Health care: UbiComp16, MobiHoc15 • Fine-grained motion detection: MobiCom15, MobiHoc16, CCS16, MobiCom14
Research Positioning 2 1 3 4
CSI to Motion • Why we can detect human motion using CSI? • Short answer: Human motion changes the value of CSI • So inversely, we can infer human motion from CSI changes • CSI Path-Phase model • Human Motion cause a dynamic multipath component • In-phase: Constructive • Out-phase: Destructive
Problem Describe • A user is making a phone call • The user’s smartphone is connect to an AP • AP collects CSI streams • AP infers user’s speaking from CSI streams
Challenges • Fine-grained motion caused CSI variance is very tiny • Easily buried in noise and interferences • Efficient CSI stream denoising methods • CSI waveforms change with context • Training per context not acceptable • Find intrinsic features in CSI dynamics correlated to fine-grained motion only
CSI Denoising • Band Pass Filter • 3-order Butterworth filter • Remove high frequency noise • Remove human breathing interference(0.2-0.33Hz) • Keep mouth movement caused frequencies(2-5Hz)
CSI Denoising • PCA Based Filtering • CSI streams of different subcarriers correlate their variations • Chose the second principal component as the filter result
Feature Extraction • Use the spectrogram for different syllables to extract the features of CSI streams • CSI-Speed model • Extract 3 contour lines of the spectrograms • Contour lines reduced computation cost • DTW to deal with different speaking speed
Test Scenario • Lip reading application • A user is making a phone call • The user’s smartphone is connected to an AP • AP collects CSI streams by sending ICMP requests • We use WiTalk to infer user’s speaking from CSI streams to verify the efficiency of our WiTalk
Test Bed • Tested on channel 36 at 5.180GHz • Context variance • Two model of smartphones • Three users • Five smartphone locations • Two AP locations • A set of 12 syllables • Repeated ten times for each context • A set of sentences from 1 word to 6 words • Repeated five times for each context
Results • 92.3% accuracy same context, 12 syllables • 82.5% accuracy mixed context, 12 syllables
Results • Sentence Detection Accuracy • 74.3% accuracy sentences up to six words
Contributions • We propose WiTalk, a feasible context-free fine-grained motion sensing solution by using WiFi CSI dynamics. • We verify the feasibility of WiTalk by applying it to the lip reading scenario
Questions Thank you!