CN112861098B

CN112861098B - Mobile phone identity verification method based on DTW algorithm and walking gait data

Info

Publication number: CN112861098B
Application number: CN202110081760.XA
Authority: CN
Inventors: 冯明旭; 刘继忠
Original assignee: Nanchang University
Current assignee: Nanchang University
Priority date: 2021-01-21
Filing date: 2021-01-21
Publication date: 2022-03-22
Anticipated expiration: 2041-01-21
Also published as: CN112861098A

Abstract

The invention discloses a mobile phone identity verification method based on a DTW algorithm and walking gait data, which comprises the steps of acquiring walking gait data of a user for a period of time in advance, then carrying out steps of preprocessing, signal segmentation, cyclic elimination of similar periods and the like to obtain a plurality of matching templates with weak similarity, generating a matching template library, then acquiring the walking gait period of the user by a signal segmentation method through the mobile phone acceleration sensor data acquired in real time, and finally calculating the similarity between the real-time data and samples in the matching template library through the DTW algorithm so as to verify whether the identity of the user is legal or not. The method has the advantages of large sample difference of the matched template, capability of reflecting the gait characteristics of the user as comprehensively as possible, solving the problems of one-sidedness, complexity and characteristic data loss of manually and manually selecting the sample by replacing a method for synthesizing the average value of the matched template, improving the identification accuracy rate and having small calculated amount and few training samples.

Description

Mobile phone identity verification method based on DTW algorithm and walking gait data

Technical Field

The invention relates to the technical field of mobile phone identity verification, in particular to a mobile phone identity verification method based on a DTW algorithm and walking gait data.

Background

With continuous upgrading and updating of various hardware carried by a smart phone, a built-in sensor of the smart phone is also continuously improved, and at present, more than ten sensors such as an accelerometer, a gyroscope, a magnetometer and a barometer are common. The use state of the user can be obtained by analyzing the sensor data, and the specific behavior or state of the user is identified, so that the identity authentication is realized, and the method and the device can be applied to various scenes such as quick unlocking of a mobile phone, quick payment and information protection.

The gait recognition is to achieve the recognition of human identity through the analysis of human or animal gait, and becomes a new biological feature recognition technology at present with the advantages of non-contact remote distance, difficult camouflage, no invasion and the like.

The Dynamic Time Warping (DTW) algorithm is a more common algorithm for calculating the similarity between Time series data. The algorithm is based on the idea of dynamic programming, the problem of template matching with different lengths is solved, fewer training samples are needed in the method, meanwhile, new actions defined by a user can be added or existing actions can be deleted more simply, the algorithm is an earlier and more classical algorithm in voice recognition, and the outstanding advantages of the algorithm are applied to other fields.

The identity verification is realized through a DTW algorithm and walking gait data, specific individual walking gait data needs to be collected firstly, a matching template is extracted, and classification and identification are carried out through calculating the similar distance between test data and the matching template, so that whether the identity is legal or not is verified. The selection of the existing matching template is usually manually completed, the gait cycle data collected by a sensor has difference due to the difference of walking step frequency, stride and road surface environment of each person, and in order to improve the universality of the matching template, two or more matching templates are usually selected manually and then synthesized into one matching template by an average value method. However, the manual method is difficult to ensure the difference among a plurality of matching templates, and if a matching template with high similarity is selected, the average synthesis effect is limited; in addition, the matching template generated by the average synthesis method causes a problem that a part of the original information is lost.

Disclosure of Invention

The invention aims to solve the defects and shortcomings of the manual selection of the matching template and the matching template average value synthesis method, and provides a method for automatically extracting walking gait cycles based on a DTW algorithm, optimizing the selection of the matching template and improving the identity verification accuracy based on the DTW algorithm, so as to solve the one-sidedness and the complexity of manual selection, and simultaneously, replacing the matching template average value synthesis method with the matching template integration method, solve the problem of characteristic data loss and improve the identification accuracy.

In order to achieve the purpose, the technical scheme adopted by the invention is as follows:

a mobile phone identity verification method based on a DTW algorithm and walking gait data comprises the following steps:

1. sample data collection and preprocessing

1.1 set the mobile phone to the state of collecting sample data, utilize built-in triaxial acceleration transducer to collect the user for a period

Interventricular walking gait data a_x(t)、a_y(t) and a_z(t)；

1.2 calculating the three-axis acceleration synthesis sequence through the collected walking gait data

1.3, preprocessing a synthetic sequence S (t):

1.3.1 normalizing the three-axis acceleration synthesis sequence S (t)

Wherein, min [ S (t)]Is the sequence S (t) minimum, max [ S (t)]Is the sequence S (t) maximum;

1.3.2 pairs of the resulting normalized data S_std(t) further processing by Gaussian smoothing iteration to obtain a sequence S_smot(t)；

2. Generating identity matching template libraries

2.1 splitting the data cycle

2.1.1 traversal sequence S_smot(t), reserving turning points and removing points with the same trend; taking the sequence S_smot(t) any of the immediately adjacent 3 numerical points is denoted as S_smot(t)、S_smot(t-1)、S_smot(t-2) if the three data points satisfy [ S ]_smot(t)-S_smot(t-1)]*[S_smot(t-1)-S_smot(t-2)]If < 0, then S_smot(t-1) is the sequence S_smot(t) a turning point, retaining and recording it to the sequence S_smot'(t)Performing the following steps; if the 3 taken data points do not meet the condition, continuously traversing along the time sequence to obtain a sequence S_smot'(t)；

2.1.2 traversal sequence S_smot' (t), reserving minimum value points and removing maximum value points; taking the sequence S_smot' (t) 2 numerical points which are arbitrarily adjacent are denoted as S_smot'(t)、S_smot' (t-1) if S is satisfied_smot'(t)-S_smot' (t-1) < 0, then S_smot' (t) is the sequence S_smot' (t) is retained and recorded to the sequence S_fit(t) in (a); if the 2 taken data points do not meet the condition, continuously traversing along the time sequence to obtain a sequence S_fit(t)；

2.1.3 traversal sequence S_fit(t), retaining minimum value points, removing maximum value points, obtaining the minimum value points which are single cycle division points of the collected walking gait data, and recording and storing position serial numbers of the cycle division points;

2.1.4S stored in 2.1.3_fit(t) mapping the position sequence numbers of the sequence period division points to the original sequence S (t), and cutting and segmenting the original sequence by taking each division point as a starting point, wherein the obtained sequence is stored and recorded as S' (t);

2.2 calculating the similar distance between each effective period and the following period in the sequence S' (t) in turn by adopting DTW algorithm, and marking as d_DTW(i, j), wherein i represents a matching template period and j represents a test set period; presetting a threshold value of the similar distance; in the first round, the first effective period in the sequence S' (t) is taken as a matching template, the subsequent period data is taken as a test group, and the similar distances d between the matching template and all the period data in the test group are sequentially calculated_DTW(1, j), j is 1,2,3 …, and if the calculated similarity distance d is greater than the predetermined similarity distance d_DTW(1, j) when the value is smaller than a set threshold value, indicating that the similarity of the two groups of periodic data is higher, and rejecting the periodic data of the test group; otherwise, reserving; the second wheel matches the periodic data still retained in the first wheel, the second effective period is used as a matching template, the subsequent periodic data is used as a test group, and the similar distance d is calculated_DTW(2, j), j is 1,2,3 …, and the rejection similarity is high, i.e. the similarity distance is less thanDetermining periodic data of a threshold value, and reserving the rest periodic data; the third wheel matches the periodic data still retained by the second wheel, the third effective period is used as a matching template, the subsequent periodic data is used as a test group, and the similar distance d is calculated_DTW(3, j), j is 1,2,3 …, eliminating cycle data with high similarity, namely the similarity distance is smaller than a set threshold value, and keeping the rest cycle data; according to the sequence, screening the periodic data in the sequence S '(t) until the data are traversed completely, and reserving the periodic data in the sequence S' (t) with the similar distance larger than the set threshold value and reserving the periodic data with the similar distance larger than the set threshold value;

2.3 extracting the reserved data cycle from the original sequence S (t), integrating to obtain the personal identity matching template library of the user, and marking as X_i(n), wherein i represents the identity number of the particular individual and n represents the total number of feature cycles in the user identity matching template library;

3. identity verification based on identity matching template library

3.1, acquiring signals of a three-axis acceleration sensor of a mobile phone of a user in real time, and taking walking gait data of every 10S as a test sample T (t);

3.2 pretreating the sample T (t) according to the method described in steps 1.2 and 1.3;

3.3 dividing the data cycle of the preprocessed sample data T' (T) according to the method in the step 2.1;

3.4 randomly selecting a set of real-time data periods T from the divided data periods_seg(t) respectively calculating the real-time data period and the user personal identity matching template library X obtained in the step 2.3 by adopting a DTW algorithm_i(n) similarity distance d for each matching template_DTW(n)；

3.5 finding d_DTWMaximum value d of (n)_maxCompared with a set period data threshold value if d_maxIf the data is less than the set threshold, the similarity of the two periodic data is high, and the identity is judged to be a legal identity; if d is_maxIf the data is not less than the threshold value, the similarity of the two periodic data is low, and the illegal identity is judged.

The invention has the beneficial effects that:

1. according to the method, a plurality of matching templates with weak similarity are obtained by collecting walking gait data of a user for a period of time, and through the steps of preprocessing, signal segmentation, cyclic elimination of a similarity period based on a DTW (delay tolerant W) algorithm and the like, as the matching templates have large difference, gait characteristics which are as comprehensive as possible can be embodied, and the one-sidedness and the complexity of manual selection are solved;

2. the invention adopts a method of matching template integration to replace a method of matching template average synthesis, solves the problem of loss of characteristic data and has high identification accuracy;

3. the method adopts the DTW algorithm in the identification process, has small calculated amount and few training samples, can automatically extract the matching template, and does not lose the characteristic information of walking gait in the whole process.

Drawings

FIG. 1 is a general work flow chart of a mobile phone identity authentication method based on DTW algorithm and walking gait data according to the invention;

FIG. 2 is a flow chart of a method for automatically extracting a feature template library of FIG. 1;

FIG. 3 is a flowchart of a method for performing identity verification by the DTW algorithm in FIG. 1;

FIG. 4 is a graph of the synthesized data of the acceleration sensor before the preprocessing in the embodiment of the present invention;

FIG. 5 is a graph of the synthesized data of the acceleration sensor after Gaussian smoothing processing according to the embodiment of the present invention;

FIG. 6 is a graph of the turning point retained after the processing of step S211 according to the embodiment of the present invention;

FIG. 7 is a graph of minimum points retained after the processing of step S212 according to the embodiment of the present invention;

FIG. 8 is a graph of a fitting function of the minimum value points retained after the processing of step S212 according to the embodiment of the present invention;

FIG. 9 is a graph of the walking gait cycle division point obtained after the processing of step S213 according to the embodiment of the invention;

FIG. 10 is a graph showing the original sequence S (t) after being cut and segmented by using the segmentation points of each period in FIG. 9 as the starting points;

FIG. 11 is a schematic diagram of the step S22 of the present invention after similar cycles are first eliminated by the DTW algorithm;

fig. 12 is a schematic diagram of a final result of the similar period elimination by the DTW algorithm in step S22 according to the embodiment of the present invention;

FIG. 13 is a partially enlarged view of FIG. 12 after similar cycles are eliminated.

Detailed Description

In order to better explain the present invention, the detailed description of the present invention is made below with reference to the accompanying drawings and examples.

Example (b): see fig. 1-13.

as shown in fig. 1, before the authentication function is started, the mobile phone authentication method based on the DTW algorithm and the walking gait data of the invention automatically extracts the walking gait data feature cycle by collecting the walking gait data of the user and using the DTW algorithm, thereby generating a corresponding integrated matching template library; after the identity authentication function is started, acquiring data of the acceleration sensor of the mobile phone in real time, acquiring a walking gait cycle by a signal segmentation algorithm, and calculating the similarity between the real-time data and a sample in a matching template library by a DTW algorithm so as to authenticate whether the identity of a mobile phone wearer is legal.

As shown in fig. 2, the process of automatically extracting the walking gait data feature cycle and generating the corresponding integrated matching template library is as follows:

s1, sample data acquisition and pretreatment

S11, setting the mobile phone to be in a sample data acquisition state, and acquiring walking gait data a of the user within a period of time by using a built-in three-axis acceleration sensor_x(t)、a_y(t) and a_z(t)；

S12, calculating a three-axis acceleration synthesis sequence through the collected walking gait data

As shown in fig. 4;

s13, preprocessing a synthesis sequence S (t):

s131, normalizing the three-axis acceleration synthesis sequence S (t)

s132, obtaining normalized data S_std(t) further processing by Gaussian smoothing iteration to obtain a sequence S_smot(t), as shown in FIG. 5;

s2, generating an identity matching template library

S21, segmenting data period

S211. traversal sequence S_smot(t), reserving turning points and removing points with the same trend; taking the sequence S_smot(t) any of the immediately adjacent 3 numerical points is denoted as S_smot(t)、S_smot(t-1)、S_smot(t-2) if the three data points satisfy [ S ]_smot(t)-S_smot(t-1)]*[S_smot(t-1)-S_smot(t-2)]If < 0, then S_smot(t-1) is the sequence S_smot(t) a turning point, retaining and recording it to the sequence S_smot' (t); if the 3 taken data points do not meet the condition, continuously traversing along the time sequence to obtain a sequence S_smot' (t), the processing results are shown in FIG. 6;

s212, traversing sequence S_smot' (t), reserving minimum value points and removing maximum value points; taking the sequence S_smot' (t) 2 numerical points which are arbitrarily adjacent are denoted as S_smot'(t)、S_smot' (t-1) if S is satisfied_smot'(t)-S_smot' (t-1) < 0, then S_smot' (t) is the sequence S_smot' (t) is retained and recorded to the sequence S_fit(t) in (a); if the 2 taken data points do not meet the condition, continuously traversing along the time sequence to obtain a sequence S_fit(t), the processing results are shown in FIG. 7; the minimum value points obtained after the fitting process are shown in fig. 8;

s213. traversing sequence S_fit(t) retaining the minimum value points, and eliminating the maximum value pointsThe minimum value point is a single cycle division point of the acquired walking gait data, the position serial number of the cycle division point is recorded and stored, and the processing result is shown in fig. 9;

s214, with S stored in step S213_fit(t) mapping the position sequence numbers of the sequence period division points to an original sequence S (t), and cutting and segmenting the original sequence by taking each division point as a starting point, wherein the obtained sequence is stored and recorded as S' (t), and the processing result is shown in figure 10;

s22, calculating the similar distance between each effective period and the subsequent period in the sequence S' (t) in sequence by using a DTW algorithm, and recording the similar distance as d_DTW(i, j), wherein i represents a matching template period and j represents a test set period; presetting a threshold value of the similar distance; in the first round, the first effective period in the sequence S' (t) is taken as a matching template, the subsequent period data is taken as a test group, and the similar distances d between the matching template and all the period data in the test group are sequentially calculated_DTW(1, j), j is 1,2,3 …, and if the calculated similarity distance d is greater than the predetermined similarity distance d_DTW(1, j) when the value is smaller than a set threshold value, indicating that the similarity of the two groups of periodic data is higher, and rejecting the periodic data of the test group; otherwise, the processing result is retained as shown in fig. 11; the second wheel matches the periodic data still retained in the first wheel, the second effective period is used as a matching template, the subsequent periodic data is used as a test group, and the similar distance d is calculated_DTW(2, j), j is 1,2,3 …, eliminating cycle data with high similarity, namely the similarity distance is smaller than a set threshold value, and keeping the rest cycle data; the third wheel matches the periodic data still retained by the second wheel, the third effective period is used as a matching template, the subsequent periodic data is used as a test group, and the similar distance d is calculated_DTW(3, j), j is 1,2,3 …, eliminating cycle data with high similarity, namely the similarity distance is smaller than a set threshold value, and keeping the rest cycle data; according to the sequence, the periodic data in the sequence S '(t) are screened until the data are all traversed, the periodic data with the similarity distance larger than the set threshold value in the sequence S' (t) are retained, the periodic data with the similarity distance larger than the set threshold value are retained, and the processing result is shown in fig. 12; a partial enlarged view of the retained cycle data is shown in fig. 13;

s23. will protectThe remaining data period is extracted from the original sequence S (t), and after integration, the personal identity matching template library of the user is obtained and recorded as X_i(n), wherein i represents the identity number of the particular individual and n represents the total number of feature cycles in the user identity matching template library.

As shown in fig. 3, the process of calculating the similarity between the real-time data and the sample in the matching template library through the DTW algorithm to verify whether the identity of the mobile phone wearer is legal includes:

s3, identity verification based on identity matching template library

S31, collecting signals of a three-axis acceleration sensor of a mobile phone of a user, and taking walking gait data of every 10S as a test sample T (t);

s32, preprocessing a sample T (t) according to the method in the steps S12 and S13;

s33, dividing the data cycle of the preprocessed sample data T' (T) according to the method in the step S21;

s34, randomly selecting a group of real-time data periods T from the segmented data periods_seg(t) respectively calculating the real-time data period and the user personal identity matching template library X obtained in the step S23 by adopting a DTW algorithm_i(n) similarity distance d for each matching template_DTW(n)；

S35, calculating d_DTWMaximum value d of (n)_maxCompared with a set period data threshold value if d_maxIf the data is less than the set threshold, the similarity of the two periodic data is high, and the identity is judged to be a legal identity; if d is_maxIf the data is not less than the threshold value, the similarity of the two periodic data is low, and the illegal identity is judged.

The above description is only for the preferred embodiment of the present invention and is not intended to limit the scope of the present invention, and all equivalent structures or equivalent transformations made by the present specification and the attached drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.

Claims

1. A mobile phone identity verification method based on a DTW algorithm and walking gait data is characterized by comprising the following steps:

1. sample data collection and preprocessing

1.1 setting the mobile phone to a sample data acquisition state, and acquiring walking gait data a of a user within a period of time by using a built-in triaxial acceleration sensor_x(t)、a_y(t) and a_z(t)；

1.3, preprocessing a synthetic sequence S (t):

1.3.1 normalizing the three-axis acceleration synthesis sequence S (t)

2. Generating identity matching template libraries

2.1 splitting the data cycle

2.1.1 traversal sequence S_smot(t) retaining turning points, and removing homodromous points to obtain a sequence S_smot'(t)；

2.1.2 traversal sequence S_smot' (t), retaining minimum value points, removing maximum value points to obtain a sequence S_fit(t)；

2.1.4S stored in 2.1.3_fit(t) mapping the sequence numbers of the sequence period division points to the original sequence S (t), and cutting and dividing the original sequence by taking each division point as a starting point to obtain a sequenceThe column store is denoted S' (t);

2.2 calculating the similar distance between each effective period and the following period in the sequence S' (t) in turn by adopting DTW algorithm, and marking as d_DTW(i, j), wherein i represents a matching template period, j represents a test group period, and period data with the similarity distance larger than a set threshold value are reserved;

3. identity verification based on identity matching template library

2. The mobile phone identity verification method based on DTW algorithm and walking gait data as claimed in claim 1, wherein the traversal sequence S in step 2.1.1_smot(t), retaining turning points, and eliminating the points with the same trend:

taking the sequence S_smot(t) any of the immediately adjacent 3 numerical points is denoted as S_smot(t)、S_smot(t-1)、S_smot(t-2) if the three data points satisfy [ S ]_smot(t)-S_smot(t-1)]*[S_smot(t-1)-S_smot(t-2)]If < 0, then S_smot(t-1) is the sequence S_smot(t) a turning point, retaining and recording it to the sequence S_smot' (t); if the 3 data points taken do not satisfy the above condition, the traversal continues in the temporal order.

3. The mobile phone identity verification method based on DTW algorithm and walking gait data as claimed in claim 1, wherein the traversal sequence S in step 2.1.2_smot' (t), retaining the minimum value point, and removing the maximum value point as follows:

taking the sequence S_smot' (t) 2 numerical points which are arbitrarily adjacent are denoted as S_smot'(t)、S_smot' (t-1) if S is satisfied_smot'(t)-S_smot' (t-1) < 0, then S_smot' (t) is the sequence S_smot' (t) is retained and recorded to the sequence S_fit(t) in (a); if the 2 data points taken do not satisfy the above condition, then the traversal continues in time order.

4. The mobile phone identity authentication method based on the DTW algorithm and the walking gait data as claimed in claim 1, wherein step 2.2 is to sequentially calculate the similar distance between each valid period in the sequence S' (t) and the following period by using the DTW algorithm, and the process of retaining the period data with the similar distance greater than the set threshold value is as follows:

s1, presetting a threshold value of a similar distance;

s2, in the first round, the first effective period in the sequence S' (t) is taken as a matching template, the subsequent period data is taken as a test group, and the similar distances d between the matching template and all the period data in the test group are sequentially calculated_DTW(1, j), j is 1,2,3 …, and if the calculated similarity distance d is greater than the predetermined similarity distance d_DTW(1, j) is less than the set threshold value, which shows that the similarity of two groups of periodic data is higher, and the test is rejectedPeriodic data of the group; otherwise, reserving;

s3, the second wheel matches the periodic data still reserved for the first wheel, the second effective period is used as a matching template, the subsequent periodic data is used as a test group, and the similar distance d is calculated_DTW(2, j), j is 1,2,3 …, eliminating cycle data with high similarity, namely the similarity distance is smaller than a set threshold value, and keeping the rest cycle data;

s4, matching the periodic data still reserved in the second wheel by using a third wheel, taking the third effective period as a matching template, taking the subsequent periodic data as a test group, and calculating the similar distance d_DTW(3, j), j is 1,2,3 …, eliminating cycle data with high similarity, namely the similarity distance is smaller than a set threshold value, and keeping the rest cycle data;

s5, screening the periodic data in the sequence S '(t) according to the sequence from S2 to S4 until the data completely traverse, and reserving the periodic data with the similar distance larger than a set threshold value in the sequence S' (t).