JP2008129704A - Personal identification device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a highly reliable personal identification device for suppressing the deterioration of authentication precision due to secular change, and for evading careless update. <P>SOLUTION: A detection part 21 detects the arm swinging action of an authenticated person as time sequential data. A similarity calculation part 24 calculates the degree of similarity of the preliminarily registered data A and input data B in an object block newly input in authenticating by using DP-matching, and a decision part 25 decides whether or not a user is the right person in question or the others according to whether or not the degree of similarity exceeds a first threshold. Also, an update part 26 prepares new update data C from the preliminarily registered registration data A based on the newly input input data B. A second threshold which is smaller than the first threshold is prepared, and when the degree of similarity exceeds the second threshold value, the decision part 25 updates the data A registered in the storage part 23 to the update data C. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

 The present invention relates to a personal authentication device based on arm swinging behavior, and more particularly to a personal authentication device that is not easily spoofed by a third party and has high safety and convenience. The present invention also relates to a personal authentication device suitable for confirming the validity of the owner of a mobile phone or the like.

  In recent years, in various situations such as in the financial industry and public institutions, there is an increasing need to perform personal authentication in order to protect bank accounts and personal information from fraud by impersonation by third parties. As a typical method for performing personal authentication, there is a method of performing personal authentication by combining a magnetic card and a password.

  However, there have been frequent third-party impersonation due to magnetic card duplication using a magnetic information duplicator and security code theft crime using a voyeur camera, and the security of the authentication method using a combination of a magnetic card and a security code has been increasing recently. Has been questioned. Further, there is a risk of losing the magnetic card or forgetting the password, which is inconvenient for the person to be authenticated.

  Based on this background, in recent years, biometric authentication using biometric information has become widespread as a personal authentication method that simultaneously realizes high safety and high convenience. The personal authentication device described in Patent Document 1 below is an authentication device based on the arm swinging behavior invented by the present inventors, and its configuration is shown in FIG.

The authentication device extracts a section to be used as an authentication target from a detection unit 11 that detects the arm swing behavior of the person to be authenticated and the entire time series data of the arm swing behavior that is output in time series by the detection unit 11. Extraction unit 12, normalization unit 13 that always normalizes a section to be used as an authentication target, and a storage unit that stores normalized arm swing behavior time-series data as registration data 14, a similarity calculation unit 15 that calculates the similarity between the time series data of the normalized arm swing action and the time series data of the arm swing action registered in advance, and the calculated similarity sets a threshold value It is comprised from the determination part 16 which determines whether it is a principal or another person according to whether it exceeds.
Japanese Patent Application No. 2006-012408

  However, as a common problem of biometric authentication, there is a problem that authentication accuracy deteriorates due to a change with time. In the authentication device based on the arm swinging behavior, a situation in which even the person cannot authenticate correctly occurs due to the change in the swing pattern over time due to the growth or decline of the arm muscles. There's a problem.

  In general, as a solution to such a problem, the method of updating the registered data by taking in the latest data at each authentication is effective, but inadvertent updating of the registered data is normal for the device. Operation should be avoided, and the burden on the user and the waiting time required for processing increase, which should be avoided.

  An object of the present invention is to provide a personal authentication device that solves the above-described conventional problems and suppresses a decrease in authentication accuracy due to secular change. Another object of the present invention is to provide a highly reliable personal authentication device that can avoid inadvertent updating. Another object of the present invention is to provide a personal authentication device that is less likely to be spoofed by a third party and is suitable for confirming the legitimacy of the owner of the mobile phone.

  In order to achieve the above-described object, the present invention provides a detection unit that detects a person's arm swing behavior as time-series data, and the entire time-series data of arm swing behavior output in time series by the detection unit. An extraction unit that extracts target section data to be used as an authentication target, a storage unit that holds the target section data as registration data A, and a registration that is pre-registered as input data B with target section data newly input at the time of authentication A similarity calculation unit that calculates the similarity with data A, a determination unit that determines whether the calculated similarity is equal to or higher than the first threshold value, and a newly entered person A first feature is that a new update data C is created from the input data B and pre-registered registration data A, and an update unit that registers the update data C in the storage unit is provided.

  According to the present invention, the determination unit has a second threshold value smaller than the first threshold value, and the similarity calculated by the similarity calculation unit has the second threshold value. There is a second feature in that a command to update the registration data A to the update data C is output when the registration data A is exceeded.

  Further, according to the present invention, the similarity calculation unit obtains the minimum path of the angle error between the sample values of the registration data A and the input data B, and the sum of the differences calculated when tracing the minimum path is similar. The third feature is that the similarity is calculated by absorbing the partial expansion / contraction and the positional deviation between the two series data due to the fluctuation by using the degree as a degree.

  Further, according to the present invention, the update unit integrates sample values of pre-registered registration data A and newly input data B corresponding to each point of the minimum path, so that the update data C There is a fourth feature in that it is created.

  In addition, the present invention has a fifth feature in that the personal authentication device is mounted on a mobile phone.

  According to the present invention, since the determination is based on the time-series data of the person to be authenticated, the characteristics of the person to be authenticated, such as the length of the arm and the structure of the muscles, and the person to be authenticated Since the unique “fake” of the media is interspersed, and the features and “fake” resulting from these structures can only be possessed by the person himself, it is difficult for a third party to impersonate and is a highly secure identity authentication device. Will be able to provide.

  In addition, according to the present invention, it is possible to update registration data that incorporates the latest pattern, so that it is possible to suppress deterioration in authentication accuracy due to secular change.

  In addition, in the apparatus described in Patent Document 1, it is necessary to normalize the sequence length, and there is a problem that although the recognition rate of the person increases, there is also a problem that an error of accepting another person is raised, but the present invention uses the DP-matching method. Therefore, it is not necessary to normalize the sequence length of the acquired data, and accuracy degradation can be suppressed.

  In addition, since the timing at which the registration data should be updated can be automatically determined, inadvertent updating of the registration data can be avoided, and the burden on the apparatus user and waiting for processing can be reduced. According to the apparatus of the present invention, it is determined whether or not the registration data should be updated only by performing authentication every time. Will be able to update.

  In addition, since the updated input data of the latest arm swing action is partly updated in the highly reliable registration data that has been registered so far, it will be updated sequentially so that it looks just like new input data that may contain noise ( Since it is not completely updated, a highly reliable personal authentication device can be provided.

  Further, when the personal authentication device according to the present invention is mounted on, for example, a mobile phone terminal or the like, it is possible to verify the legitimacy of whether or not the user of the mobile phone is really the correct owner through arm swinging behavior. It becomes like this.

  Hereinafter, the present invention will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing a configuration of an embodiment of a personal authentication apparatus of the present invention.

  The personal authentication device uses a detection unit 21 that detects the person's arm swing behavior as time-series data, and uses the entire time-series data of the arm swing behavior output in time series by the detection unit 21 as an authentication target. An extraction unit 22 that extracts target section data, a storage unit 23 that stores the target section data as registration data A, and registration data A that is registered in advance as input data B with the target section data newly input at the time of authentication. Based on the similarity calculation unit 24 that calculates the similarity, the determination unit 25 that determines whether the calculated similarity is greater than or equal to the threshold value, and the input data B that is newly input And an update unit 26 that creates new update data C from registered data A registered in advance. The hardware configuration of the present invention can be the same as that shown in FIG.

Below, the said detection part 21, the extraction part 22, the memory | storage part 23, the similarity calculation part 24, the determination part 25, and the update part 26 are demonstrated in detail.
○ Detection unit 21

The detection unit 21 can be realized using, for example, an acceleration sensor, a gyro sensor, or the like. When the acceleration sensor is used as the detection unit 21, the acceleration of the arm swing behavior is detected, and when the gyro sensor is used, the angular acceleration is detected. As an example, FIG. 3 shows time-series changes in acceleration when the arm swinging action shown in FIG. 2 is performed. The horizontal axis represents time, and the vertical axis represents the acceleration value. In the above example, acceleration sensors are installed in each of the x-axis, y-axis, and z-axis directions, and time-series changes in acceleration in the x-axis, y-axis, and z-axis directions are recorded as time-series data of arm swing behavior. It is a result.
○ Extraction unit 22

  Next, the processing of the extraction unit 22 will be described with reference to FIG. 4 (acceleration in the y-axis direction) as an example. The extraction unit 22 inputs the value U output in time series by the detection unit 21 (FIG. 4 (a)), and (1) only the section used for authentication from the entire time series data of arm swing behavior. Is identified (FIG. 4B), and (2) the corresponding section is extracted (FIG. 4C).

  First, an operation (FIG. 4 (b)) for specifying only a section to be used as an authentication target from among records of the entire time-series data of arm swing behavior will be described.

  The section used as an authentication target is from when the arm swinging action starts to when it ends. In order to determine the start point and the end point, the variance of the latest few samples (for example, 10 samples) is sequentially calculated with respect to the time series data.

  It can be considered that the more the arm swinging action is, the larger the values of the acceleration, the angular acceleration, and the speed mentioned in the previous example are as the output values of the detection unit 21, and the variance is larger. On the other hand, when the arm swinging behavior is almost stationary, the values of acceleration, angular acceleration, and velocity are all small and the variance is also small.

  Therefore, when the variance of the most recent samples of the time series data is larger than the set threshold value T, it is considered that the arm swinging action is large, and it is determined that the arm is being shaken. Conversely, if the variance of the most recent samples is smaller than the set threshold value T, the arm swinging action is small, especially when the variance shows a value close to 0, It is determined that it is not shaken.

As described above, the start point t ₀ and the end point t ₁ of the section to be used as a desired authentication target can be specified by the threshold value determination with respect to the variance value of the most recent samples. Further, by adjusting the size of the threshold value T, the section length Δ used as an authentication target can be adjusted.

Next, a recording section corresponding to the section Δ used as an authentication target specified by the above-described processing is extracted from the entire recording (see FIG. 4C).
○ Storage unit 23

Next, each of the two switching means 27 and 28 of the device of the present invention shown in FIG. 1 is connected to the contact S1 in the storage unit 23, and a legitimate owner of the mobile phone terminal holds the mobile phone terminal and swings his arm. When the action is performed, the target section data acquired through the processing of the above-described extraction unit 22 is registered as registration data of the arm swinging action of the legitimate owner of the terminal. On the other hand, when the authentication is performed, each of the switching means 27 and 28 is connected to the contact S2, and the registered data of the legitimate owner registered in the storage unit 23 is read out.
○ Similarity calculation unit 24

  Next, the similarity calculation unit 24 will be described. The similarity calculation unit 24 calculates the similarity between the registration data A, which is time-series data of arm swing behaviors registered in advance in the storage unit 23, and the input data B, which is newly acquired arm swing behavior data.

Since the data is obtained using a triaxial acceleration sensor composed of the x-axis, y-axis, and z-axis, the sample point t _A (= 1, 2, 3,... Shown in FIG. The sample value a _tA of the registered data A in (, m) is a vector composed of three components having components of the x-axis, y-axis, and z-axis, respectively, and the point t _B (b) shown in FIG. = 1, 2, 3,..., N), the sample value b _tB of the input data B is also a vector composed of three components having components on the x-axis, y-axis, and z-axis. 5, m and n are sequence lengths, a ₁ , a ₂ , a ₃ ,...; B ₁ , b ₂ , b ₃ ,... Are sample values of registered data A and input data B, respectively. .

Defined by the following equation, the sample value of the registration data A at the sample point t _A a _tA and obtained from the sample values b _tB of the input data B at the sample point t _B angle error θ (a _tA, b _tB) from the following formula The minimum path is obtained by DP-matching, which can calculate the degree of similarity by absorbing the partial expansion and contraction between the two series data due to the fluctuation and the positional deviation, and the sum Diff of the angle errors corresponding to the minimum path is set as the degree of similarity.

_{θ (a tA, b tB)} = ABS (ArcCos {(a tA, b tB) / (| a tA || b tB |)})

Here, ABS is an absolute value, (a _tA, b _tB) is a _tA and b _tB inner product, | it means each magnitude of a _tA and b _tB | a _tA | and | b _{tB. Further, 0 ≦ θ (a tA,} b tB) is ≦ [pi.

  DP-matching is a similarity calculation method that is frequently used in speech recognition and the like, and associates the elements of each series data between two series data, and the two series data due to fluctuations This is a method for calculating the degree of difference by absorbing partial expansion and contraction and misalignment.

  The calculation method of the total angle Diff of DP errors by DP-matching and the minimum path at that time will be described using the simple examples shown in FIGS. The sequence length of the registration data A is m, and the sequence length of the input data B is n.

First, the angle error θ (a _tA, b _tB) at each sample point by the above equations is calculated. Assume that the results are as shown in the table of FIG. Here, in the j-th row and i-th column of the table, the sample value a _{i at} the sample point t _A = i of the registered data A and the sample value b _j of the input data B at the point t _B = j of the input data B are obtained. Angular error θ (a _i , b _j ) is calculated and entered.

  When the sum of angle errors is calculated without taking into account partial expansion and contraction and positional deviation between the two series data, the values of i and j are simultaneously incremented by 1 in the registration data and the input data, that is, from the first row and the first column. The sum is calculated in the lower right direction. In DP-matching, i is incremented by 1 with only registered data in consideration of partial expansion and contraction and misalignment, that is, the sum is obtained in the right direction in the table, or j is incremented by 1 with only input data, that is, a table. The calculation is performed while allowing the sum to be calculated in the downward direction, and the path having the smallest possible angle error is found.

  However, when similarity calculation is performed in consideration of partial expansion / contraction and positional deviation between two series data, that is, when only one of i or j is incremented, partial expansion / contraction between two series data and Add a penalty for considering misalignment. As summarized in FIG. 7, a penalty = 0 is imposed in the lower right direction, a penalty = P is imposed in the right direction, and a penalty = P is imposed also in the lower direction. P is set as 0 ≦ P ≦ π.

In this way, as shown in the Z diagram of FIG. 7 and the hatched cells of FIG. 8, a path having the minimum sum of angle errors is found. 7 is a diagram in which the upper left portion is extracted from FIG. 6, and FIG. 8 is a diagram showing the shortest path and the update data C (= c1, c2, c3,..., Cl). The square path indicates an example of the path that minimizes the angle error (shortest path) with respect to FIG. 6, and the update data C will be described later.
○ Determining unit 25

Next, the determination unit 25 determines whether or not the person is the person based on the similarity calculated by the similarity calculation unit 24. It is considered that the similarity between the registration data and the authentication data is higher as the total angle Diff of DP-matching is smaller. In particular, Diff = 0 means that the registration data and the authentication data completely match. On the contrary, it is considered that the similarity between the registration data and the authentication data is lower as the total angle error Diff is larger. Therefore, the determination threshold value D1 is set, and if the value of Diff is equal to or less than the set threshold value D1, it is determined that the person is the person, and if the value is greater than D1, the person is not the person.
○ Update unit 26

As shown in FIG. 8, it is assumed that the minimum cost path by DP-matching is a path indicated by hatched cells. The sequence length of the update data C is equal to the number l of hatched cells. The sample value a _tc of the update data C at the sample point t _C is integrated by averaging the sample values of the registration data A and the input data B at the corresponding sample points in the table.

For example, in the case of FIG. 8, c ₃ is in the third row and the second column, so it is obtained by taking the sum of a ₂ and b ₃ and dividing by 2. Moreover, since c l _-1 is in the m-1 column (n-1) -th row is determined by dividing by 2 taking the sum of the a _m-1 and b _n-1.
Others are as follows.
c ₁ = {a ₁ + b ₁ } / 2 c ₂ = {a ₂ + b ₂ } / 2 c ₃ = {a ₂ + b ₃ } / 2
_{c 4 = {a 3 + b} 4} / 2 c 5 = {a 4 + b 4} / 2 c 6 = {a 5 + b 4} / 2
_{c 7 = {a 6 + b} 4} / 2 c 8 = {a 7 + b 4} / 2 c 9 = {a 8 + b 4} / 2
...
cl _-5 = {am _-3 + bn _-3 } / 2 cl _-4 = {am _-3 + bn _-2 } / 2 cl _-3 = {am _-3 + _{bn- 1} } / 2
c _l−2 = {a _m−2 + b _n−1 } / 2 c _l−1 = {a _m−1 + b _n−1 } / 2 c _l = {a _m + b _n } / 2

The update data C integrated in this way (= c ₁ , c ₂ , c ₃ ,..., C _l ) is re-registered in the storage unit 23.

  Next, the timing for re-registering the update data C created by the update unit 26 in the storage unit 23 will be described.

  As a problem common to biometric authentication, there is a problem that authentication accuracy deteriorates due to changes over time. This is due to the growth and decline of arm muscles, and the pattern of swinging changes over time, resulting in a situation where even the person himself cannot be authenticated correctly.

  The following two methods can be considered as timing for executing the update. One is a method for forcibly performing an update for each authentication, and the other is a method for determining whether an update is necessary for each authentication and automatically performing an update if necessary. It is.

  In order to realize the former updating method, each of the switching means 27 and 28 is connected to the contact S2, and the switching means 29 is connected to the contact S5. As a result, the data in the storage unit 23 is forcibly updated each time authentication is performed. In order to prevent updating during authentication, each of the switching means 27 and 28 is connected to the contact S2, and the switching means 29 is connected to the contact S3.

  On the other hand, there are two methods for realizing the latter updating method. To realize these methods, each of the switching means 27 and 28 is connected to the contact S2, and the switching means 29 is connected to the contact S4. To do.

  In the first method, when the determination unit 25 authenticates that the user is the person, a command is issued from the determination unit 25 to turn on the switching unit 20.

  In the second method, another threshold value D2 for update determination different from the threshold value D1 for personal authentication determination set in the determination unit 25 is provided. Note that D2 is (0 ≦) D2 <D1.

  In other words, the update is performed when the total angle Diff of DP-matching between the input data B and the registration data A is equal to or less than D1 and D2 <Diff ≦ D1. If Diff is less than or equal to D2, what is authenticated as the principal is not updated.

  Based on the above update determination, the determination unit 25 controls the switching means 20. That is, if Diff ≦ D2 and D1 <Diff, the switching means 20 is turned off, and if D2 <Diff ≦ D1, the switching means 20 is turned on.

  According to this method, it is considered that the degree of similarity with respect to previously registered data gradually decreases due to changes in arm muscles over time, but registration data is updated while the decrease in similarity is small. Is not performed, and is updated when the voltage drops to some extent, that is, when the voltage drops to D2 or more.

  In this embodiment, it is possible to automatically determine when to update the registration data, and the update data C is created by integrating the previously stored registration data and the new input data. Update of data can be avoided and reliability can be improved.

  The personal authentication device of the present invention having the above-described configuration is mounted on, for example, a mobile phone terminal, and verifies the validity of whether or not the user of the mobile phone is really the right owner through arm swinging behavior. Will be able to.

It is a block diagram which shows the schematic structure of one Embodiment of this invention. It is a figure which shows an example of an arm swing action. It is a graph which shows an example of the time series data (change of an angular acceleration) of the arm swing action obtained when an acceleration sensor is used as a detection part. It is explanatory drawing of the procedure which extracts the data of the area used as the object of authentication from the time series data of arm swing action. It is explanatory drawing of the sample point of the registration data A and the input data B. FIG. It is explanatory drawing of DP-matching. It is explanatory drawing of the minimum path | route of the angle error of each sample value of the registration data A and the input data B in DP-matching. It is explanatory drawing of the update data of the minimum path | route of the angle error in DP-matching. It is a block diagram which shows the structure of the outline of the conventional personal authentication apparatus.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 21 ... Detection part, 22 ... Extraction part, 23 ... Memory | storage part, 24 ... Similarity calculation part, 25 ... Determination part, 26 ... Update part.

Claims (5)

A detection unit that detects the person's swing motion as time-series data;
An extraction unit that extracts target section data to be used as an authentication target from the entire time-series data of the arm swing action that is output in time series by the detection unit;
A storage unit that holds the target section data as registered data A;
A similarity calculation unit for calculating a similarity with registration data A registered in advance as input data B with the target section data newly input at the time of authentication;
A determination unit that determines whether the calculated similarity is greater than or equal to a first threshold value, whether the person is the other person or not,
An updater for creating new update data C from newly input data B and preregistered registration data A, and for registering the update data C in the storage unit Authentication device. In the personal identification device according to claim 1,
The determination unit has a second threshold value smaller than the first threshold value, and when the similarity calculated by the similarity calculation unit exceeds the second threshold value, A personal authentication device that outputs a command to update registration data A to update data C. In the personal authentication device according to claim 1 or 2,
The similarity calculation unit obtains the minimum path of the angle error of each sample value of the registration data A and the input data B, and uses the sum of the differences calculated when tracing the minimum path as the similarity. A personal authentication device characterized in that a similarity is calculated by absorbing partial expansion and contraction between two series data due to fluctuation and a positional deviation. In the personal identification device according to claim 3,
The update unit creates the update data C by integrating pre-registered registration data A and newly input sample data B corresponding to each point of the minimum path. A personal authentication device. In the personal identification device according to any one of claims 1 to 4,
A personal authentication apparatus, wherein the personal authentication apparatus is mounted on a mobile phone.