Gait authentication system and method thereof
View Patent ↗A gait authentication system and method is provided. The system includes a mobile computing device configured to be carried by a person while moving with a unique gait. A first sensor (e.g. an accelerometer) is carried by the mobile computing device and generates a first signal. A second sensor (e.g. a gyroscope) is carried by the mobile computing device and generates a second signal. A gait dynamics logic implements an identity vector (i-vector) approach to learn feature representations from a sequence of arbitrary feature vectors carried by the first and second signals. In the method for real time gait authentication, the computing of invariant gait representations are robust to sensor placement while preserving highly discriminative temporal and spatial gait dynamics and context.
1. A gait authentication system comprising:
a mobile computing device configured to be carried by a person while moving with a unique gait;
a first sensor carried by the mobile computing device and generating a first signal;
a second sensor carried by the mobile computing device and generating a second signal;
gait dynamics logic implementing an identity vector (i-vector) approach to learn feature representations from a sequence of arbitrary feature vectors carried by the first and second signals;
a gait dynamic image (GDI) generated from the feature representations; and
extraction logic to extract gait specific signatures unique to the moving person from observed sensory data in the first and second signals generated from the first and second sensors, wherein the first and second signals include a plurality of irrelevant data.
2. The gait authentication system of claim 1 , further comprising:
a plurality of GDIs, wherein the GDIs are invariant relative to a physical orientation of the first and second sensors and adapted to present consistent images before and after repositioning the mobile device.
3. The gait authentication system of claim 1 , further comprising:
a gait signature representing locomotion pattern of the person moving with a unique gait;
wherein the gait signature characterizes the unique gait of the person invariantly relative to position of the first and second sensors.
4. The gait authentication system of claim 1 , further comprising:
an independent relationship between the first signal and the second signal, wherein the gait dynamics logic captures gait dynamics in each gait cycle of the moving person the independent relationship between the first and second signals.
5. The gait authentication system of claim 1 , further comprising:
learning logic to learn the feature representations from the sequence of arbitrary feature vectors.
6. The gait authentication system of claim 5 , further comprising:
computation logic using the feature representations learned from the learning logic to calculate at least one of (i) vector distance similarity measurements, (ii) input measurements to an additional feature transform modelling, and (iii) input measurements to an additional machine learning modelling.
7. The gait authentication system of claim 6 , further comprising:
modeling logic to build a universal background model ( 0 ) by pooling at least some of the feature vectors from a training data set and to compute a supervector for each authentication GDI feature sequence.
8. The gait authentication system of claim 7 , further comprising curve estimation logic to fit smooth horizontal curves across one GDI to maximize overall intensity.
9. The gait authentication system of claim 8 , wherein any one of the modeling logic, estimation logic, computation logic, and the gait dynamics logic analyzes a supervector for each enrollment or authentication gait GDI feature sequence of L frames {y 1 , y 2 , . . . , y L }, where each frame is a feature vector of dimension F; and
the posterior probability (N c ) and Baum-Welch statistics ({tilde over (F)} c ) for each Gaussian component are computed as:
N c =Σ t=1 L P ( c|y t ,Ω), (Equation 1)
and
{tilde over (F)} c =Σ t=1 L P ( c|y t ,Ω)( y t −m c ), (Equation 2)
where m c is mean vector for Gaussian component c; and
a supervector M is obtained by concatenating {tilde over (F)} c for all Gaussian components to form a vector of fixed dimension C·F for an input sequence of arbitrary length L.
10. The gait authentication system of claim 9 , wherein the same logic implementing Equation 1 and Equation 2 conducts factor analysis in the supervector space using a simplified linear model:
M=m+Tw (Equation 3)
where m is a subject independent component, T is a low rank rectangular matrix, and w is the i-vector.
11. The gait authentication system of claim 10 , wherein the same logic implementing Equation 1, Equation 2, and Equation 3 computes an i-vector as:
w =( I+T t Σ −1 NT ) −1 T t Σ −1 M, (Equation 4)
where N is a diagonal matrix consisting of diagonal blocks of N c I.
12. The gait authentication system of claim 11 , wherein once the i-vector is extracted for each gait sequence, the computation logic computes a similarity between two gait sequences as the cosine distance between two corresponding i-vectors according to:
d
(
w
1
,
w
2
)
=
<
w
1
,
w
2
>
w
1
w
2
.
(
Equation
5
)
13. A method comprising the steps of:
providing a mobile computing device carrying a first sensor and a second sensor, wherein the first sensor is an accelerometer and the second sensor is a gyroscope;
generating a first signal with the first sensor during a unique gait of a moving person;
generating a second signal with the second sensor during the unique gait of the moving person; and
computing invariant gait representations with an identity vector (i-vector) approach, wherein the gait representations are robust to the first sensor placement and the second sensor placement and preserve discriminative temporal and spatial gait dynamics;
authenticating the moving person, based at least in part on, the invariant gait representation computed through an i-vector approach;
creating a first gait dynamics image (GDI) from acceleration measurements obtained in the first sensor;
creating a second GDI from rotational measurement obtained in the second sensor; and
learning a GDI feature sequence for later gait identity vector extraction.
14. The method of claim 13 , wherein the step of computing invariant gait representations with an i-vector approach further comprises the step of:
computing a gait specific feature, based at least in part on, pairwise interactions inside each gait cycle.
15. The method of claim 14 , wherein the step of computing invariant gait representations with an i-vector approach further comprises the step of:
encoding dynamics for a time series and encoding dynamics for local interactions in the GDI.
16. The method of claim 13 , further comprising the step of:
extracting signatures of the unique gait of the moving person from sensory data corrupted with variations from various irrelevant sources.
17. The method of claim 16 , further comprising the step of:
building a universal background model (UBM) using a Gaussian mixture model (GMM) by pooling all or a subset of a set feature vectors from a training data set.
18. The method of claim 17 , further comprising the step of:
computing a supervector for each enrollment or authentication gait GDI feature sequence.
19. The method of claim 18 , further comprising the step of:
conducting factor analysis for the supervector utilizing a simplified linear mode.
20. The method of claim 19 , further comprising the step of:
computing the similarity between two gait sequences as the cosine distance between their corresponding i-vectors.
21. The method of claim 20 , wherein the step of authenticating is accomplished by applying cosine similarity scores of i-vectors against a set of known indexes to authenticate the moving person.