Method for recognizing identity and gesture based on radar signals

Patent metadata

Field	Value
Publication number	US-11947002-B2
Application number	US-202117218174-A
Country	US
Kind code	B2
Filing date	Mar 31, 2021
Priority date	Apr 9, 2020
Publication date	Apr 2, 2024
Grant date	Apr 2, 2024

Abstract

A method for recognizing an identity and a gesture based on radar signals includes: reading a radar echo signal reflected by various gestures of a tester and received by a radar sensor; frequency-mixing the radar echo signal with a radar transmission signal; after filtering and centralizing a frequency-mixed signal, training and obtaining a neural network capable of identity verification and a neural network capable of gesture recognition; in a real-time detection process, verifying an identity of a user; and if the identity is verified, determining that a gesture of the user is valid; verifying the gesture of the user and executing a corresponding operation according to correspondence between the gesture of the user and an operation. It can be determined by the method whether the gesture belongs to the corresponding user and whether the operation corresponding to the gesture is performed according to the determined result.

First claim

What is claimed is:

1. A method for recognizing an identity and a gesture based on radar signals, including following steps:
Step 1, reading a radar echo signal reflected by various gestures of a tester and received by a radar sensor;
Step 2, frequency-mixing the radar echo signal with a radar transmission signal;
Step 3, filtering a frequency-mixed signal by a high-pass filter;
Step 4, performing a centralization operation on filtered signal data;
Step 5, determining a neural network model suitable for features of the radar signal, training the model by using a preprocessed signal in the Step 4 and an identity information label of the tester, to obtain a neural network A for identity verification;
Step 6, obtaining distance information, speed information and angle information, according to the preprocessed signal in the Step 4 and parameters of the radar sensor, to calculate three-dimensional coordinates and Doppler values of a moving target;
Step 7, dividing space above a radar in a space grid method; mapping a spatial area into a three-dimensional matrix; determining a grid position where the moving target is located in the space, according to the three-dimensional coordinates obtained in the Step 6; accumulating the Doppler values obtained in the Step 6 at a matrix unit corresponding to the grid position, as a size of an element of the matrix unit; training a designed neural network by the matrix and a gesture information label, to obtain a neural network B for gesture recognition;
Step 8, performing a constant false alarm rate detection on the radar echo signal detected in real time, to determine whether a gesture recognition is performed; and if the gesture recognition is performed, proceeding to the Step 9; otherwise continuing to wait for detection;
Step 9, performing a preprocessing of the Steps 2 to 4 on the detected radar echo signal; sending the processed signal to the neural network A trained in the Step 5, to extract features for analysis; verifying an identity of a user through identity information contained in the features; and if the identity is verified, determining that the gesture of the user is valid and proceeding to Step 10, otherwise continuing to wait for detection; and
Step 10, performing a processing of the Steps 6 and 7 on the data preprocessed in the Steps 2 to 4, and inputting the processed data to the neural network B, to determine a gesture of a user and to perform a corresponding operation according to a correspondence between the gesture of the user and an operation,
wherein in the Step 6, the distance information R is obtained through a fast time dimension FFT, according to the preprocessed signal in the Step 4 and the parameters of the radar sensor:

f
movingBeat

≈

f

s
⁢
t
⁢
a
⁢
t
⁢
i
⁢
c
⁢
B
⁢
e
⁢
a
⁢
t

=

2
⁢

f
c

⁢
R

C
⁢

t
c

R
=

Ct
c

2
⁢

f
c

×

f

s
⁢
t
⁢
a
⁢
t
⁢
i
⁢
c
⁢
B
⁢
e
⁢
a
⁢
t

;

the speed information v is obtained through a slow time dimension FFT:

f
d

=

2
⁢
f
⁢
v

C

⁢

⁢

v
=

f
d

⁢
C

2
⁢
f

,

wherein f movingBeat and f staticBeat are frequencies of beat signals when a target is in motion and static states respectively, f d is a Doppler frequency, f c a span, R is a distance to the target, C is a speed of light, t c is a frequency sweep period, f is a center frequency of a Chirp signal, and v is a speed of the target; and
the angle information θ is obtained according to a plurality of transmitting and receiving antennas of the radar sensor:

Δ
⁢
Φ

=

2
⁢
πΔ
⁢

⁢
d

λ

⁢

⁢

θ
=

sin

-
1

⁡

(

λ
⁢
Δ
⁢
Φ

2
⁢
π
⁢

Assignees

• Univ Zhejiang

Inventors

• Wang Yong
• Chen Junyi
• Cao Jiahe

Classifications

• G06N3/09

  Supervised learning · CPC title

• G01S13/89 Primary

  for mapping or imaging · CPC title

• G01S13/505

  using Doppler effect for determining closest range to a target or corresponding time, e.g. miss-distance indicator · CPC title

• G01S13/53

  performing filtering on a single spectral line and associated with one or more range gates with a phase detector or a frequency mixer to extract the Doppler information, e.g. pulse Doppler radar {(G01S13/5244 takes precedence)} · CPC title

• G06F3/017

  Gesture based interaction, e.g. based on a set of recognized hand gestures (interaction based on gestures traced on a digitiser G06F3/04883) · CPC title