Advertisement

Identification and Authentication Using Blink with Smart Glasses

Sep. 30, 2022
Advertisement

More Related Content

More from sugiuralab(20)

Advertisement

Identification and Authentication Using Blink with Smart Glasses

  1. Identification and Authentication Using Blink with Smart Glasses K e i o U n i v e r s i t y Yohei Kawasaki and Yuta Sugiura The SICE Annual Conference 2022
  2. • Widespread of Smart Glasses • Google Glass • JINS MEME • Growing market size • Need for Identification and Authentication 2 [1] Google「Glass Enter Edition2」, https://www.nttdocomo.co.jp/biz/product/xr/googleglass/, (Accessed August 19th, 2022) [2] JINS MEME,https://www.jins.com/jp/item/WCE-19S-001.html, (Accessed August 19th, 2022) Widespread of Smart Glasses Google Glass[1] JINS MEME[2]
  3. 3 Type of Authentication Passwords, PINs IC cards, Keys Knowledge-Based Authentication Possession-Based Authentication Biometric Authentication Irises, Fingerprint Gait, Handwriting Physical Characteristics Behavioral Characteristics
  4. • Using blink as a behavioral feature 4 Blink as Behavior for Authentication
  5. • With a head-mounted device 5 Related Work: Identification and Authentication Method with Head Mounted Device • Using blink Acoustic properties of the nose [Isobe, 2021] Acoustic properties of the skull [Schneegass, 2016] Electro-Encephalo-Gram (EEG) [Abo-Zahhad, 2014] Blink pattern [Westeyn, 2004] Head shape [Fujii, 2021]
  6. • Detecting blink by photo-reflectors mounted on smart glasses for Identification and Authentication • Contactless • Low cost • Small size • Hands-free 6 Method
  7. 7 Implementation: Hardware Get sensor value with photo-reflectors Transmit to computer via Arduino UNO Analysis
  8. • Identification • Predicts which class the input data belong to • Evaluated by accuracy • Authentication • Predicts whether the input data is the person or not • Evaluated by • FRR (False Rejection Rate) • FAR (False Acceptance Rate) • EER (Equal Error Rate) 8 Identification and Authentication
  9. • Estimates the class with the smallest DTW distance as the correct class. • Participants: 𝐴𝑘 (𝑘 = 1, 2, … 𝑁) 9 Identification Method
  10. • Consider the authentication of 𝐴𝑘 as an example • Participants: 𝐴𝑘 (𝑘 = 1, 2, … 𝑁) • Calculate Mahalanobis distance to only class k data 10 Authentication Method
  11. • 𝐹𝑅𝑅 = 𝐹𝑎𝑙𝑠𝑒 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑇𝑟𝑢𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒+𝐹𝑎𝑙𝑠𝑒 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒 • 𝐹𝐴𝑅 = 𝐹𝑎𝑙𝑠𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑇𝑟𝑢𝑒 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒+𝐹𝑎𝑙𝑠𝑒 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒 11 Authentication Method θ: threshold
  12. • Set a threshold for accepting that it is the person. • FRR (False Rejection Rate) • FAR (False Acceptance Rate) • EER (Equal Error Rate) 12 Authentication Method EER
  13. • Prompts participants to blink in 2 patterns. • Pattern 1: Blink 5 times in a row at any time. • Pattern 2: Blink 5 times in a row at a tempo of 120 BPM. • 20 measurements each • Frame rate: 70fps • 10-fold cross validation with 140 data 13 Number of participants 7 Sex 5 male, 2 female Mean of Age 26.6 SD of Age 12.4 Experiment Participants information Photo-reflectors
  14. 14 Result: Identification Blink pattern Accuracy (%) Pattern 1 88.6 Pattern 2 97.1 Each accuracy Confusion matrix of pattern 1 Confusion matrix of pattern 2
  15. 15 Result: Authentication Blink pattern EER (%) Pattern 1 8.6 Pattern 2 6.4 Each EER FRR and FAR of pattern 1 FRR and FAR of pattern 2
  16. Limitation • Sensor value changes depending on sunlight • Setting the mounting position Future Work • Expansion of the range of participants • Develop robustness • Change of mounting position • Measurements of different days 16 Limitation and Future Work
  17. 17 Summary Background Identification and Authentication with Smart Glasses Related work Identification and Authentication with head mounted device and use of blink Suggestion Use blink as Behavioral Characteristics Implementation Sensing blink with photo reflector Evaluation 10-fold cross validation Identification: accuracy Authentication: FRR, FAR and EER Result Identification: 97.1 % Authentication: 6.4 % Limitations Mounting position, effects of sunlight Future works Change of mounting position and Measurements of different days to develop robustness
Advertisement