Smart user authentication through actuation of daily activities leveraging wifi-enabled research direction

Published in Mobihoc ' 17, is the result of research by Chen Yingying team.

Objective: To authenticate users through their daily actions (Walking,stationary activities).
Main contributions: 1 through the relationship between wireless signal and action, to obtain a unique human body characteristics and behavioral characteristics, and then through day-to-day action to authenticate the user. 2 use CSI's amplitude and phase to describe the user's unique action. 3 The use of depth learning methods to detect the user's daily movements, and prevent illegal user attacks (this point is relatively good) 4 in two types of experimental environment (lab and apartment) to obtain the authentication accuracy of 94% and 91% respectively.
Price: 1 A pair of equipment

2) 5 months ' time

STE (short): Short-time Energy

Effect Display:

For a crest, if multiple peaks, this method will have some problems. I'm still thinking about whether I can continue to improve and improve the accuracy and stability of the starting point of the detection action.

Subcarrier selection:
To eliminate the "negative effects caused by" unstable subcarriers, a covariance based the function is scoring To assess each subcarrier ' s correlation level with its neighboring subcarriers as follows:

K: Indicates the number of adjacent subcarrier

I,j: Represents a different subcarrier

The effect is as follows:

My thoughts: I have always tended to retain sensitive subcarrier and remove the stable subcarrier. The work of this paper is to remove the unstable subcarrier and keep the stable subcarrier. This is very interesting, cause me to think. Which one is closer to the real theory? It also needs to be explored and analysed.




My personal point of view: The work is still not out of our common solution to the action recognition techniques and methods, but only a new combination of ideas, so innovative or not too prominent. "We show that the proposed system is resilient to spoofing attacks when integrating the feature abstractions from the Dee P Learning model with the SVM classifier "