TWI617999B - Double identification door access system and method thereof - Google Patents

Abstract

A dual identification access control system and method thereof are characterized in that the image capturing unit of the portable smart mobile device captures a three-dimensional image with user biometrics and gesture recognition for identity recognition to open the door lock and allow a specific user to enter. The biometrics of the user's 3D image are regarded as biometric keys and transmitted by the mobile communication system for the first stage identification. After the conditions are met, the data is transmitted to the gating device where the position is located, and the gesture recognition is regarded as a private key. In the second stage of identification, the second stage will meet the conditions before the door will be released for the user to enter.

Description

Double identification access control system and method thereof

The invention relates to an identification access control identification system and a method thereof, and an access control system exclusively using three-dimensional (3D) image interactive double identification.

At present, the access control system mostly uses face-held cameras to perform face recognition, and its face recognition is a method that is not necessarily safe to adopt two-dimensional identification. Because it is easy to obtain photos of others and confuse the system of two-dimensional identification methods, it is impossible to use It is quite inconvenient to identify the use on the road.

A similar concept is mentioned in the following prior patent documents: the biometric door lock system of Taiwan Patent I560353, the authentication method of the access control of Taiwan Patent I355624, and the mobile electronic device and access control system using the same, Taiwan Patent Application TW200409045 A biometric access control system and a multi-identification access control method of Taiwan Patent I476734.

It can be seen that there are still many shortcomings in the above-mentioned methods of use, which is not a good design, but needs to be improved.

In view of this, the object of the present invention is to provide an access control system and a method for dual recognition of a 3D image interactive identification device for an access control system.

The double-identified access control system that achieves the above object of the invention requires at least the following steps. The image capturing unit on the portable smart mobile device captures the 3D image biometrics of any part of the user, and transmits the biometric data to the background host via the mobile communication network. After the background management host compares successfully, the user's The private key data is sent to the identification controller via the network, and the user enters through the private key to open the access control.

The image capturing unit of the portable smart mobile device arbitrarily captures the 3D image with the biometric characteristics of the user to open the biological key, and after the verification is passed, the private key comparison is opened, and the verification can be performed to release the access control. A two-layer verification mechanism for the access switch is achieved.

The above described features and advantages of the invention will be apparent from the following description.

As shown in FIG. 1 , a system architecture diagram of an access control system identified by a smart mobile device according to an embodiment of the present invention includes: a portable smart mobile device 1, a background management host 2, an identification controller 3, Private key 4.

The portable smart mobile device 1 can be a mobile device, a tablet computer and the like having a network function and an image capturing function, and its main function is to capture through an image capturing unit (for example, a camera, a camera, etc.). The user's biometric 3D image is transmitted to the background management host 2 via the mobile communication network (for example, supporting 3rd generation, 4th generation mobile communication (3G, 4G), etc.) via the mobile communication network. The biometric feature is obtained by the image capturing unit of the portable smart mobile device 1 and is uniquely identifiable by the user, and is not limited to a face or a fingerprint, and is adjusted according to the needs of the embodiments of the present invention.

The background management host 2 can be a desktop computer, a computer workstation or a server. The background management host 2 includes a biometric identification module 11 and a private key distribution module 12 as shown in FIG. 2, and the background management host 2 can have a central processing unit (CPU), a microprocessor (Microprocessor), and the like. A digital signal processor (DSP), a programmable controller, or the like executes the software modules, and has a communication unit corresponding to the portable smart mobile device 1 to obtain the data transmitted by the portable smart mobile device 1.

The identification controller 3 can be configured with a controller that can obtain the private key 4. In the embodiment, the private key 4 is a gesture recognition data, and the recognition controller 3 is configured with an image capturing unit such as a camera or a camera to obtain a gesture. The image has a specific processor to compare the acquired gesture image with the data of the private key 4. It should be noted that the background management host 2 and the identification controller 3 respectively have corresponding wired or wireless communication units (for example, WiFi, Ethernet, etc.) to transfer data to each other (for example, private key 4, comparison) Results, etc.).

When the user travels to the access control area, the portable smart mobile device 1 can be hand-held and the biometric 3D image data can be retrieved through the image capturing unit. The biometric feature capture can be automatically imaged by the hand or handheld portable smart mobile device 1 to adjust to the best recognized image to improve the recognition rate. These image data will be transferred to the background management host 2.

The background management host 2 of the embodiment of the present invention uses a Local Binary Pattern (LBP) operation as a main algorithm, and the method can be divided into two steps: eigenvalue calculation and statistical operation.

The eigenvalue calculation is to subtract each pixel value in the image from the pixel value by 3×3 neighborhood, and if the result is negative, the value is marked as 0, and if it is positive, it is marked as 1. Next, start clockwise with a binary code starting from the right side of the center point. The calculation method is as follows: (1-1): ...(1-1) The reference point is adjacent to all pixel values, Is the reference point pixel value, Is the number of adjacent pixels.

The statistical operation divides the image into a number of blocks, performs statistical operations on the feature values of each block, presents the data in a histogram, and then aggregates the histograms of all blocks into one piece of data. In the embodiment of the present invention, the face image and the depth image (that is, the 3D image data) are cut into nine equal parts, and the LBP feature value is normalized for each aliquot, and then collected into a piece of data to calculate the completed LBP. Values are also stored in the database for comparison purposes. The background management host 2 compares and identifies the image data of the user according to the established data, that is, compares and identifies the 3D image data. If the user's data has been established in the database and the comparison is successful, the background management host 2 sends the user's private key data to the identification controller 3 via the wired or wireless network.

The background management host 2 and the database are compared equations such as (1-2): j = 1,2,3...M ...(1-2) where, For the LBP eigenvalues in the database, For the LBP feature value to be identified. N represents the feature dimension, and M represents the stroke of the comparison in the database. The background management host 2 compares the face depth information with the information in the built-in database, and compares the error values one by one to find the biometric with the smallest error.

It should be noted that there are many techniques for biometric comparison. Those who apply the embodiments of the present invention can adopt other kinds of comparison techniques according to requirements, and the present invention is not limited to the LBP operation.

When the background management host 2 receives the biometric data of the user, it determines whether it is a legitimate user via the biometric identification module 11 (software) operation, and if it is confirmed as a legitimate user, the private key distribution module 12 will be private. The key data (for example, an identification code) is sent to the identification controller 3 for the user to open the private key 4 (such as the identification controller 3 and the background management host 2 exchange messages), so that the identification controller 3 has the function of judging the opening of the door. (Control to release the access control). After the identification controller 3 is used, the data having the function of judging the opening of the door will be automatically deleted, so that the private key data cannot be stolen, and the private key data for judging the opening of the door is reloaded by the background management host 2 when the user is in the lower position. . When the second stage (biometric and private key comparison) is matched, the recognition controller 3 will unlock the door and let the user pass.

In this embodiment, the private key 4 is gesture recognition, and the gesture recognition method may be to find both the feature point and the straight line. These feature points are the use of endpoints and turning points: first define eight orientations and a 19x19 square box. If only one orientation is found, it is the endpoint. If there are two orientations and the orientation difference is four, it is a straight line. If the orientation difference is not Four are turning points.

The method of finding a straight path defines n as the number of linear feature points, thereby obtaining the number of possible straight paths: ...(1-3)

At the same time, you can also find the angle of each line, and then substitute each pixel in the image into (1-4): ...(1-4)

By weighting each possible straight path, you can know whether the line exists or not. Finally, according to the comparison between the number of lines and the angle, the action of opening the door can be performed.

The embodiment of the invention utilizes the neural network to have a high degree of ability to learn and deal with nonlinear problems, and establishes an access control system for identifying the front smart mobile device to provide an accurate identification function. The most basic component of a neural network is shown in Figure 3. The neural network architecture 21 is shown. After the input value X and the weight value W are input to the neuron, the neuron starts to be calculated. The calculation formula is (1-5). formula: ...(1-5) where The neuron weight value of the nervous system, Feature input value, The partial weight of the neuron, The output value of the neuron, the number of n- series inputs, and the j- positive integer.

The embodiment of the present invention uses a neural network architecture and a program to construct a reverse neural network. The implementation of this mode includes learning algorithms, hidden layer neurons, hidden layers, learning rate, and hidden layers. Six items, such as the output layer conversion function. The learning algorithm is divided into 8 steps, which are explained as follows:

Step 1: Determine the number of layers in the network and the number of neurons in each layer. It is assumed here that the architecture of the network is an input layer, a hidden layer and an output layer, that is, the network is assumed to be a three-layer network architecture, and the number of neurons in the input layer is assumed to be The number of neurons in the hidden layer And the number of neurons in the output layer has One.

Step 2: Set the initial weighted value and initial bias value of the network by the random distribution random number. Because the neurons in different layers are connected to each other, if The weight of the i- th neuron in the input layer and the h- th neuron in the hidden layer, due to Input neurons and The hidden neurons, so you can use a double layer loop to set the initial weight between all input layers and hidden layers, as follows: for i = 1 to For h = 1 to Average distribution random random number =

Similarly, if we make For the h-th neuron and j-th weight of the neural weights between the hidden layer output layer element, the method of setting the initial values of the weights between the hidden layer and the output layer for all the following: for h = 1 to For j = 1 to Average distribution random random number =

If you set the initial bias value in the network, you need to pay attention to the fact that only the hidden layer and the output layer have the bias value, and the input layer does not. In fact, the input layer is not computationally capable, it simply outputs the signals received by one neuron in parallel to the neurons in the hidden layer. If For the h-th partial weights of the hidden layer neurons, For the bias value of the jth neuron of the output layer, the method of setting the initial bias value is as follows: for h = 1 to Average distribution random random number = For j = 1 to Average distribution random random number =

Step 3: Enter the training sample , ,..., And target output value , ,..., . input value , ,..., It can be any real value, but since the inverse transfer network uses the sigmoid function as the nonlinear transfer function of the neuron, the value range of the network's inferential output value must also fall in [0,1]. Between, so the target output value , ,..., Its range must also fall between [0, 1].

Step 4: Calculate the inference output value of the network , ,..., . The calculation method is to first calculate the output value of the hidden layer, as follows: for h = 1 to ...(1-6) for h. =1 to ...(1-7) where A weighted product of the hidden layer neurons and h, and To hide the output value of the hth neuron of the layer, the weighted product sum to be collected Do another nonlinear conversion.

Since the input signal of the output layer is from the output value of the hidden layer, its inference output value can be calculated as follows: for j = 1 to ...(1-9) for j = 1 to ...(1-10) where and They are the weighted product sum and inference output values of the jth neuron in the output layer.

Step 5: Calculate the difference between the output layer and the hidden layer. The formula for calculating the gap in the output layer is as follows: for j = 1 to ... (1-11) where Is the amount of gap in the jth neuron of the output layer. In formula (1-11) Indicates the error between the target output value and the network inference output value, so Express versus The measure of error between.

The formula for calculating the amount of hidden layer gap is as follows: for h = 1 to ...(1-12) where Indicates the amount of gap in the hth neuron of the hidden layer. Note that there is a subform in (1-12): This formula represents the weighted product sum of the output layer gaps. and so The calculation is related to the amount of gap in the output layer, which means that the error of the output layer is reversed to the hidden layer to calculate the amount of the gap.

Step 6: Calculate the correction amount of the weight value and the correction amount of the bias value between the layers. If It represents the h-th neuron and j-th weighted value of the correction amount between the hidden layer output layer neural element, and let To indicate the correction value of the bias value of the jth neuron, the method of calculating all the weighting values and the correction value of the bias value is as follows: for h =1 to For j =1 to ...(1-13) ...(1-14) where For the learning rate, the value is generally 0.1~1.0. In order to speed up the convergence of the network, formulas (1-13) and (1-14) can be rewritten into ...(1-15) ...(1-16) where For the inertia factor, the value is generally 0.0~0.9.

Similarly, if The weighted value correction amount between the i- th neuron of the input layer and the h- th neuron of the hidden layer, and The formula for the correction value of the amount of h-th partial weights of the hidden layer neurons, and the weighting values of all partial therebetween weight correction amount is calculated as follows: for i = 1 to For h = 1 to ...(1-17) for h = 1 to ...(1-18)

Step 7: Update the weighted values and the partial weights between the layers. The method of updating the weighted value between the hidden layer and the output layer and the output layer bias value is as follows: for h =1 to For j =1 to ...(1-19) for j = 1 to

Similarly, the method of updating the weighted value between the input layer and the hidden layer and the hidden layer bias value is as follows: for i = 1 to For h = 1 to ...(1-20) for h = 1 to

Step 8: Repeat steps 3 through 7 until the network converges. The learning process is usually performed one training sample at a time, until the network learns all the training samples, called a learning circle, allowing the network to repeatedly learn a learning loop until the network converges. To test whether the network is converging, define the following error function to represent the learning quality of the network: ...( 1-21) This formula represents the sum of the squared errors of the individual neurons in the output layer. Because in the learning process, we want the inference output value of the network And target output value The closer the better, the calculated value of (1-21) should be less than a reasonable range. The number of hidden layer neurons is obtained by the averaging method. The hidden layer, learning rate, hidden layer and output layer conversion functions all find the best value by erroneous method.

It should be noted that the private key in this embodiment adopts a gesture recognition technology. However, in other embodiments, the private key may be a specific identification code, a specific pattern, or the like, and the application is adjusted according to the requirements of the present application.

Features and effects

The access control system and the method for identifying the smart mobile device of the present invention use the portable smart mobile device 1 to capture 3D images with user biometrics and transmit them through a mobile communication network for biometric identification and system Identification is faster.

2. The access control system and the method for identifying the smart mobile device of the present invention can arbitrarily capture 3D images with biometric characteristics of the user, so that the system can be versatile.

3. The access control system and method for identifying the smart mobile device of the present invention have two-factor authentication to make the system more secure.

Although the present invention has been disclosed in the above embodiments, it is not intended to limit the present invention, and any one of ordinary skill in the art can make some changes and refinements without departing from the spirit and scope of the present invention. The scope of the invention is defined by the scope of the appended claims.

1‧‧‧Portable Smart Mobile Device

2‧‧‧Background management host

3‧‧‧ Identification controller

4‧‧‧Private keys

11‧‧‧Biometric Identification Module

12‧‧‧Private key distribution module

21‧‧‧ class neural network architecture

1 is a system architecture diagram of an access control system for identifying a portable smart mobile device according to an embodiment of the present invention; FIG. 2 is a structural diagram of a console management host according to an embodiment of the present invention; FIG. 3 is a diagram of a host management host according to an embodiment of the present invention; Neural network architecture like this.

Claims (6)

A dual-identification access control system includes: a portable smart mobile device that captures a three-dimensional image having user biometrics and transmits the three-dimensional image through a mobile communication network; and a background management host, comprising: a biometric identification module, Receiving biometric data in the three-dimensional image and determining whether it is a legitimate user; and a private key distribution module, if the biometric identification module is confirmed as a legitimate user, the private key distribution module sends a private key data; and an identification controller, obtaining the private key data, and obtaining an external private key through the image capturing unit, and comparing the private key data with the external private key based on gesture recognition The number of straight lines and the angle of the line, thereby controlling the release of the access control, wherein the private key is related to a gesture image, and the recognition controller defines eight orientations. If the gesture image has two orientations among the orientations and the orientation is different by four, The recognition controller determines that it is a straight line. The double identification access control system according to claim 1, wherein the identification controller and the background management host mutually transmit a message, so that the identification controller has a function of determining to open the door, and the identification controller uses the private key data. After the completion, the data having the function of judging the opening of the door is deleted, so that the private key data cannot be stolen, and the private key data for judging the opening of the door is reloaded by the background management host when the user is in the lower position. The dual identification access control system of claim 1, wherein the background management host uses a neural network-like architecture to identify biometrics in the three-dimensional image. A dual identification access control method includes: capturing a three-dimensional image having a user biometric through a portable smart mobile device and transmitting the three-dimensional image through a mobile communication network; and receiving biometric data in the three-dimensional image through a background management host And determining whether the user is a legitimate user; if the biometric identification module is confirmed as a legitimate user, sending a private key data to the identification controller through the background management host; and capturing the image through the identification controller The unit obtains an external private key, wherein the private key is related to a gesture image; and the recognition key controller compares the private key data with the external private key according to the gesture recognition by a straight line angle and a linear angle, thereby controlling The access control is released, wherein the identification controller defines eight orientations. If the gesture image has two orientations among the orientations and the orientation is different by four, the recognition controller determines that the line is a straight line. The double identification access control method according to claim 4, wherein the step of comparing the private key data with the external private key through the identification controller comprises: using the private key data to complete the door opening The function data is deleted, so that the private key data cannot be stolen; and when the lower user is to be re-loaded by the background management host, the private opening is judged. There is key information. The dual identification access method according to claim 4, wherein the step of receiving the biometric data in the three-dimensional image through the background management host and determining whether it is a legitimate user comprises: using the neural network in the background management host The network architecture identifies biometrics in the three-dimensional image.