CN107833328B - Access control verification method and device based on face recognition and computing equipment - Google Patents

Abstract

The invention discloses a door control verification method and device based on face recognition and a computing device, wherein the method comprises the following steps: acquiring an image shot by a camera; inputting the image into a second neural network to obtain a face recognition result; the second neural network is obtained by utilizing output data of at least one middle layer of a first neural network trained in advance to conduct guiding training, wherein the number of layers of the first neural network is more than that of the second neural network; judging whether the access control verification is successful or not by using the face recognition result; if so, completing the access control verification and opening the access control for the identified object. According to the invention, the trained neural network with fewer layers is used for quickly and accurately calculating to obtain the face recognition result, whether the access control verification is successful or not is conveniently judged according to the obtained face recognition result, and after the access control verification is completed, the access control is opened for the recognition object, so that the time for the recognition object to wait for the opening of the access control is reduced, and the user experience of the recognition object on the access control verification is improved.

Description

Access control verification method and device based on face recognition, and computing equipment

technical field

The invention relates to the field of deep learning, in particular to a face recognition-based access control verification method and device, and computing equipment.

Background technique

With the development of modern technology, electronic access control has been widely used. The control of the entrance through electronic access control verification can effectively restrict personnel from entering the controlled area and achieve the purpose of ensuring the safety of the controlled area. Such as electronic card access control verification, fingerprint access control verification, face recognition access control verification, etc. The response of electronic card swiping access control verification is not sensitive enough, and sometimes it is necessary to swipe the card multiple times to verify the success; fingerprint access control verification requires high fingerprint clarity, which leads to the need to input fingerprints multiple times to verify successfully; face recognition is reliable and smarter ,safer.

In the prior art, a neural network is used to detect an image acquired by a camera during face recognition. However, the commonly used neural network often has multiple intermediate layers, which can obtain accurate face recognition results, but the calculation speed of the multi-layer intermediate layers will be slow, which cannot quickly detect the image and feedback the face recognition results in time. , making the access control verification slow. When a neural network with fewer intermediate layers is used, due to the small number of intermediate layers, the calculation speed is faster, the face recognition results can be quickly fed back, and the speed of access control verification is improved. However, limited by the number of layers, it may cause problems such as limited computing power, poor fitting ability, and inaccurate results.

SUMMARY OF THE INVENTION

In view of the above problems, the present invention is proposed to provide a face recognition-based access control verification method, device, and computing device that overcome the above problems or at least partially solve the above problems.

According to one aspect of the present invention, a face recognition-based access control verification method is provided, which includes:

Get the image captured by the camera;

Input the image into the second neural network to obtain the face recognition result; wherein, the second neural network uses the output data of at least one intermediate layer of the pre-trained first neural network to guide the training to obtain, and the layer of the first neural network is obtained. The number is more than the number of layers of the second neural network;

Use the face recognition result to judge whether the access control verification is successful;

If so, complete the access control verification and open the access control for the identified object.

Optionally, the access verification includes entry verification and/or exit verification.

Optionally, the method includes:

Using the face recognition result, combined with the time when the face recognition result is obtained, the behavior track of the recognized object is recorded.

Optionally, completing the access control verification and opening the access control to the identified object further includes:

Obtain the personal information of the recognized object according to the face recognition result; wherein, the personal information includes permission setting information for the recognized object;

According to the personal information of the identification object, open the access control corresponding to its permission setting information for the identification object.

Optionally, the method further includes:

If the access control verification fails, the image is recorded and an alarm message is issued.

Optionally, the training process of the second neural network includes:

Input the training sample data of face recognition into the first neural network obtained by training, and obtain the output data of at least one layer of the first intermediate layer of the first neural network;

Input the training sample data of face recognition into the second neural network to be trained, and obtain the output data and final output data of at least one second intermediate layer of the second neural network, at least one second intermediate layer and at least one second intermediate layer. The first intermediate layer of the layer has a corresponding relationship;

The second neural network is trained using the loss between the output data of the at least one second intermediate layer and the output data of the at least one first intermediate layer, and the loss between the final output data and the pre-labeled output data.

Optionally, at least one layer of the first intermediate layer includes the bottleneck layer of the first neural network; at least one layer of the second intermediate layer includes the bottleneck layer of the second neural network.

Optionally, using the loss between the output data of at least one second intermediate layer and the output data of at least one first intermediate layer, and the loss between the final output data and the pre-labeled output data, the second neural The network is trained further including:

The weight parameters of the second neural network are updated according to the loss between the output data of at least one second intermediate layer and the output data of at least one first intermediate layer, and updated according to the loss between the final output data and the pre-labeled output data The weight parameter of the second neural network is used to train the second neural network.

Optionally, before inputting the input data of the training sample into the second neural network to be trained, before obtaining the output data of at least one second intermediate layer of the second neural network and the final output data, the method further includes:

The training sample data for face recognition is down-sampled, and the processed data is used as the training sample data for face recognition of the second neural network.

Optionally, using the loss between the output data of at least one second intermediate layer and the output data of at least one first intermediate layer, and the loss between the final output data and the pre-labeled output data, the second neural The network is trained further including:

Utilize the loss between the output data of at least one second intermediate layer and the output data of at least one first intermediate layer, as well as the final output data and the pre-labeled output data of the training sample data of face recognition after downsampling processing The loss between is used to train the second neural network.

Optionally, the method further includes:

The images captured by the camera are collected as input data for training samples of face recognition, and the images are manually labeled, and the labeled images are used as pre-labeled output data.

According to another aspect of the present invention, a face recognition-based access control verification device is provided, which includes:

an acquisition module, suitable for acquiring images captured by the camera;

The recognition module is suitable for inputting the image into the second neural network to obtain the face recognition result; wherein, the second neural network uses the pre-trained output data of at least one intermediate layer of the first neural network to guide and train, and the first neural network is obtained. The number of layers of one neural network is more than the number of layers of the second neural network;

The judgment module is suitable for judging whether the access control verification is successful by using the face recognition result;

Open the module, if it is suitable, complete the access control verification, and open the access control for the identified object.

Optionally, the access verification includes entry verification and/or exit verification.

Optionally, the device further includes:

The recording module is suitable for recording the behavior track of the recognized object by using the face recognition result in combination with the time when the face recognition result is obtained.

Optionally, the opening module is further adapted to:

According to the face recognition result, the personal information of the recognition object is obtained; wherein, the personal information includes the permission setting information for the recognition object; according to the personal information of the recognition object, the access control corresponding to the permission setting information is opened for the recognition object.

Optionally, the device further includes:

The alarm module is suitable for recording images and sending alarm information if access control verification fails.

Optionally, the device further includes: a face recognition network guidance training module;

The face recognition network guidance training module includes:

a first output unit, adapted to input the training sample data of face recognition into the first neural network obtained by training, and obtain the output data of at least one layer of the first intermediate layer of the first neural network;

The second output unit is adapted to input the training sample data of face recognition into the second neural network to be trained, and obtain the output data and final output data of at least one second intermediate layer of the second neural network, at least one layer The second intermediate layer has a corresponding relationship with at least one first intermediate layer;

The guidance training unit is adapted to utilize the loss between the output data of the at least one second intermediate layer and the output data of the at least one first intermediate layer, and the loss between the final output data and the pre-labeled output data, for the first Two neural networks are trained.

Optionally, at least one layer of the first intermediate layer includes the bottleneck layer of the first neural network; at least one layer of the second intermediate layer includes the bottleneck layer of the second neural network.

Optionally, the instruction training unit is further adapted to:

The weight parameters of the second neural network are updated according to the loss between the output data of at least one second intermediate layer and the output data of at least one first intermediate layer, and updated according to the loss between the final output data and the pre-labeled output data The weight parameter of the second neural network is used to train the second neural network.

Optionally, the face recognition network guidance training module further includes:

The down-sampling unit is adapted to perform down-sampling processing on the training sample data for face recognition, and use the processed data as the training sample data for face recognition of the second neural network.

Optionally, the instruction training unit is further adapted to:

Utilize the loss between the output data of at least one second intermediate layer and the output data of at least one first intermediate layer, as well as the final output data and the pre-labeled output data of the training sample data of face recognition after downsampling processing The loss between is used to train the second neural network.

Optionally, the device further includes:

The collection module is suitable for collecting images captured by the camera as input data of training samples for face recognition, and manually labeling the images, and using the labeled images as pre-labeled output data.

According to another aspect of the present invention, a computing device is provided, including: a processor, a memory, a communication interface, and a communication bus, and the processor, the memory, and the communication interface communicate with each other through the communication bus;

The memory is used for storing at least one executable instruction, and the executable instruction enables the processor to perform the operations corresponding to the above-mentioned face recognition-based access control verification method.

According to yet another aspect of the present invention, a computer storage medium is provided, the storage medium stores at least one executable instruction, and the executable instruction causes the processor to perform operations corresponding to the above-mentioned face recognition-based access control verification method.

According to the face recognition-based access control verification method, device, and computing device provided by the present invention, the image captured by the camera is obtained; the image is input into the second neural network to obtain the face recognition result; wherein, the second neural network uses pre-training The output data of at least one intermediate layer of the first neural network is obtained by guiding the training, and the number of layers of the first neural network is more than the number of layers of the second neural network; use the face recognition result to determine whether the access control verification is successful; if so, complete Access control verification and open access control to identified objects. In the present invention, the output data of at least one intermediate layer of the first neural network with a higher number of layers is used to guide the training of the second neural network with a smaller number of layers, so that the second neural network obtained by training keeps its fast calculation condition. , greatly improving its accuracy. The second neural network can be used to quickly and accurately calculate the face recognition result, which is convenient for judging whether the access control verification is successful according to the obtained face recognition result. time, and improve the user experience of the identification object for access control verification.

The above description is only an overview of the technical solutions of the present invention, in order to be able to understand the technical means of the present invention more clearly, it can be implemented according to the content of the description, and in order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand , the following specific embodiments of the present invention are given.

Description of drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are for the purpose of illustrating preferred embodiments only and are not to be considered limiting of the invention. Also, the same components are denoted by the same reference numerals throughout the drawings. In the attached image:

Fig. 1 shows the flow chart of the access control verification method based on face recognition according to an embodiment of the present invention;

FIG. 2 shows a flowchart of a face recognition network guidance training method according to another embodiment of the present invention;

Fig. 3 shows the flow chart of the access control verification method based on face recognition according to another embodiment of the present invention;

4 shows a functional block diagram of an access control verification device based on face recognition according to an embodiment of the present invention;

5 shows a functional block diagram of an access control verification device based on face recognition according to another embodiment of the present invention;

FIG. 6 shows a schematic structural diagram of a computing device according to an embodiment of the present invention.

Detailed ways

Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited by the embodiments set forth herein. Rather, these embodiments are provided so that the present disclosure will be more thoroughly understood, and will fully convey the scope of the present disclosure to those skilled in the art.

FIG. 1 shows a flowchart of an access control verification method based on face recognition according to an embodiment of the present invention. As shown in Figure 1, the access control verification method based on face recognition specifically includes the following steps:

In step S101, an image captured by a camera is acquired.

The camera can capture real-time monitoring images, such as installing cameras at the entrance of the community, garage entrance, elevator room, company and other places, it is very convenient to view the real-time situation. The image captured by the camera is acquired. In this embodiment, the person in the image is identified. Therefore, the image including the person captured by the camera is acquired, so that the image can be processed subsequently.

Step S102, the image is input into the second neural network to obtain a face recognition result.

The second neural network is a shallow neural network, which has a small number of layers and a fast calculation speed, and is generally suitable for mobile devices, small calculators and other devices. The number of layers of the first neural network is greater than the number of layers of the second neural network. The accuracy of the first neural network is higher. Therefore, the pre-trained output data of at least one intermediate layer of the first neural network is used to guide the training of the second neural network, so that the final output data of the second neural network is the same as that of the first neural network. The final output data of the neural network is consistent, and the computing performance of the second neural network is greatly improved on the premise of retaining the computing speed of the second neural network. The second neural network is obtained by guiding training using the output data of at least one intermediate layer of the pre-trained first neural network, wherein the samples used in the training of the first neural network and the second neural network are both training samples for object recognition.

Input the image into the second neural network to obtain the face recognition result. Wherein, the face recognition result may be the recognition result of the frontal face or the side face of a certain angle of the person in the image.

Step S103, using the face recognition result to determine whether the access control verification is successful.

Using the face recognition result, if the face recognition result is the recognition object A, find out whether there is A in the user information stored in the access control database, if yes, then judge that the access control verification is successful, and execute step S104; if not, the access control verification fails, no Identify the object to open the door. Or the face recognition result is the recognition object A, find the recognition object A from the user information stored in the access control database, and the information of the recognition object A is on-the-job, then judge that the access control verification is successful, and execute step S104; the information of the recognition object A is resignation, then The access control verification fails, and the access control is not opened for the identified object.

In step S104, the access control verification is completed, and the access control is opened for the identified object.

The access control verification process is completed, the access control verification is successful, and the access control is opened for the recognized object identified by the face in the image.

Further, if there are multiple people in the image, there are multiple face recognition results. Use multiple face recognition results to determine whether the access control verification is successful, and only open the access control for one or more identification objects that have completed the access control verification successfully, and close the access control for the identified users who fail the access control verification. Enter trailing.

According to the access control verification method based on face recognition provided by the present invention, the image captured by the camera is obtained; the image is input into the second neural network to obtain the face recognition result; wherein, the second neural network uses the pre-trained first neural network The output data of at least one middle layer of the first neural network is obtained by guiding the training, and the number of layers of the first neural network is more than the number of layers of the second neural network; the face recognition result is used to judge whether the access control verification is successful; if so, the access control verification is completed, and the Identify the object to open the door. In the present invention, the output data of at least one intermediate layer of the first neural network with a higher number of layers is used to guide the training of the second neural network with a smaller number of layers, so that the second neural network obtained by training keeps its fast calculation condition. , greatly improving its accuracy. The second neural network can be used to quickly and accurately calculate the face recognition result, which is convenient for judging whether the access control verification is successful according to the obtained face recognition result. time, and improve the user experience of the identification object for access control verification.

FIG. 2 shows a schematic flowchart of a face recognition network guidance training method according to an embodiment of the present invention. As shown in FIG. 2 , the guidance training step of the face recognition network includes the following steps:

Step S201 , input the training sample data of face recognition into the first neural network obtained by training, and obtain the output data of at least one first intermediate layer of the first neural network.

The first neural network is a pre-trained and solidified neural network. Specifically, the first neural network has been trained using multiple training sample data for face recognition in advance, and the first neural network has been well suited for face recognition. Among them, the first neural network preferably uses a deep neural network, such as a neural network applied to a cloud server, which has good performance, large amount of calculation, high accuracy and slow speed. The first neural network can output the output data of the first intermediate layer of multiple layers. For example, the first neural network contains 4 layers of the first intermediate layer, which are the first intermediate layer of the fourth layer, the first intermediate layer of the third layer, and the first intermediate layer of the second layer. The first intermediate layer and the first intermediate layer are the first intermediate layer, wherein the first intermediate layer of the first layer is the bottleneck layer of the first neural network.

The training sample data of face recognition is input into the first neural network, and the output data of at least one layer of the first intermediate layer of the first neural network can be obtained. Here, only the output data of the first intermediate layer of one layer can be obtained, the output data of the first intermediate layer of adjacent layers can also be obtained, or the output data of the first intermediate layer of the multiple layers separated from each other can be obtained. According to the actual situation, it is not limited here.

Step S202 , input the training sample data of face recognition into the second neural network to be trained, and obtain output data and final output data of at least one second intermediate layer of the second neural network.

The second neural network is the neural network to be trained in the guidance training of the face recognition network, which is a shallow neural network, such as a neural network applied to a mobile terminal, which has limited computing power and poor performance. The first neural network has more layers than the second neural network. For example, the number of layers of the first neural network is 4, which are the first middle layer of the fourth layer, the first middle layer of the third layer, the first middle layer of the second layer and the first middle layer of the first layer; the second neural network The number of layers is 2, which are the second intermediate layer of the second layer and the second intermediate layer of the first layer.

The training sample data of face recognition is input into the second neural network, and the output data of at least one second intermediate layer of the second neural network is obtained. Wherein, at least one second intermediate layer has a corresponding relationship with at least one first intermediate layer. For example, the first intermediate layer of the first neural network corresponds to the first intermediate layer of the second neural network, and the second intermediate layer of the first neural network corresponds to the second intermediate layer of the second neural network. The two middle layers correspond.

The obtained output data of the second middle layer of the second neural network needs to correspond to the obtained output data of the first middle layer of the first neural network. If the output data of the two first middle layers of the first neural network are obtained, It is also necessary to obtain the output data of the two second intermediate layers of the second neural network. For example, to obtain the output data of the first layer of the first neural network and the first intermediate layer of the second layer, correspondingly obtain the output data of the first layer of the second neural network and the second intermediate layer of the second layer.

Preferably, at least one layer of the first intermediate layer may include the bottleneck layer of the first neural network, that is, the first layer of the first intermediate layer of the first neural network, and at least one layer of the second intermediate layer may include the bottleneck layer of the second neural network, That is, the first layer and the second intermediate layer of the second neural network. The bottleneck layer is the highest layer of the hidden layer in the neural network, and the middle layer with the least dimension of the output vector. Using the bottleneck layer can ensure that the final output data is more accurate during subsequent training, and better training results can be obtained.

In inputting the training sample data of face recognition into the second neural network to be trained, in addition to obtaining the output data of at least one second intermediate layer of the second neural network, it is also necessary to obtain the final output data of the second neural network , so as to use the final output data to calculate the loss and train the second neural network.

Considering that the second neural network is a shallow neural network, when the training sample data for face recognition is large, directly using the training sample data for face recognition will affect the operation speed of the second neural network. Optionally, the training sample data for face recognition can be down-sampled first. For example, when the training sample data for face recognition is a picture, the down-sampling process can first reduce the resolution of the picture, and the processed face recognition data can be processed by down-sampling. The training sample data is used as the training sample data for face recognition input by the second neural network. In this process, the second neural network uses the low-resolution face recognition training sample data after downsampling for training, the first neural network uses the high-resolution face recognition training sample data for training, and uses two neural networks for training. When the output data of the network is trained, the second neural network can also obtain a high-resolution output result for the low-resolution face recognition training sample data.

Step S203, using the loss between the output data of at least one second intermediate layer and the output data of at least one first intermediate layer, and the loss between the final output data and the pre-labeled output data, to the second neural network. to train.

According to the loss between the output data of at least one second intermediate layer and the output data of at least one first intermediate layer, the weight parameters of the second neural network can be updated, so that the second neural network has at least one second intermediate layer. The output data is as close as possible to the output data of at least one first intermediate layer of the first neural network.

At the same time, according to the loss between the final output data of the second neural network and the pre-labeled output data, the weight parameters of the second neural network can be updated to make the final output data of the second neural network as close to the pre-labeled output data as possible , to ensure the accuracy of the final output data of the second neural network. In the above manner, the training of the second neural network is completed. Optionally, when the second network uses the training sample data of face recognition after downsampling, it is also necessary to pre-mark the training sample data of face recognition after downsampling, so as to obtain the face recognition after downsampling. The pre-labeled output data of the training sample data. According to the loss between the final output data of the second neural network and the pre-labeled output data after downsampling, the weight parameters of the second neural network can be updated to make the final output data of the second neural network as close to downsampling as possible. The pre-labeled output data of the processed data ensures the accuracy of the final output data of the second neural network.

According to the face recognition network guidance training method provided by the present invention, the training sample data of face recognition is input into the first neural network obtained by training, and the output data of at least one layer of the first intermediate layer of the first neural network is obtained; Input the training sample data of face recognition into the second neural network to be trained, and obtain the output data and final output data of at least one second intermediate layer of the second neural network, at least one second intermediate layer and at least one second intermediate layer. Layer the first intermediate layer has a corresponding relationship; using the loss between the output data of at least one second intermediate layer and the output data of at least one first intermediate layer, and the loss between the final output data and the pre-labeled output data , to train the second neural network. By using the output data of at least one layer of the first intermediate layer of the first neural network to train the output data of at least one layer of the second intermediate layer corresponding to the second neural network, the second neural network can be maintained at a constant computational cost. In this case, the performance of the second neural network is greatly improved, the training time for training the second neural network is effectively reduced, and the training efficiency of the second network is improved.

FIG. 3 shows a flowchart of a method for authentication of access control based on face recognition according to another embodiment of the present invention. As shown in Figure 3, the access control verification method based on face recognition specifically includes the following steps:

Step S301, acquiring an image captured by a camera.

Step S302, the image is input into the second neural network to obtain a face recognition result.

The above steps refer to steps S101-S102 in the embodiment of FIG. 1 , and details are not described herein again.

Step S303, using the face recognition result to determine whether the access control verification is successful.

The face recognition result is used to match the user information stored in the access control database. If the match is successful, it is determined that the access control verification is successful, and step S304 is performed; if the match fails, the access control verification fails, and step S308 is performed. When matching, further attention needs to be paid to the current state of user information. If the access control verification is the company access control verification, the matching will only match the user information stored in the access control database whose current status is the in-service status, and not match the user information in the resignation status; or the access verification is the community access verification verification, only when matching The current state of the user information stored in the access control database is the user information of the current resident, and it does not match the user information that has been moved away.

Step S304, according to the face recognition result, obtain the personal information of the recognized object.

Step S305, according to the personal information of the identification object, open the access control corresponding to the authority setting information for the identification object.

After the access control verification is successful, according to the face recognition result, the personal information of the recognized object is further obtained from the access control database. Among them, the personal information includes permission setting information for the identification object. Specifically, the personal information of the identification object includes that he is a resident on the 5th floor of Building 1 of the community, and the permission setting information is that he can use the elevator of Building 1, but only the elevator buttons on the 5th floor can be used. Using the face recognition results, the elevator of Building 1 successfully authenticated the access control of the identified object, and the elevator door of the elevator of Building 1 was opened. Then according to the obtained personal information of the identification object, especially the permission setting information, the corresponding elevator button on the 5th floor can be automatically lighted, so that the identification object can directly reach the fifth floor of Building 1; or, it can be manually pressed on the identification object. When the elevator button on the 5th floor is pressed, the elevator button on the 5th floor will be lit. However, when the identification object manually presses the elevator buttons of other floors, the elevator buttons of other floors will not light up, and the identification object is not allowed to use the elevator buttons of other floors other than the permission setting information, so as to ensure the internal safety of the residents of the community, and control the community as much as possible. Potential hidden problems within the household.

In step S306, the behavior track of the recognized object is recorded using the face recognition result in combination with the time when the face recognition result is obtained.

Using the face recognition result and the time when the face recognition result is obtained during recognition, the time is associated with the recognition object, and the behavior track of the recognition object at each time can be recorded. According to the behavior trajectory, the required information is automatically obtained. Such as commuting card information, daily behavior habits, etc. Specifically, such as company access control verification, the face recognition result of the recognition object A is obtained at 9 o'clock, and the face recognition result of the recognition object A is obtained at 18 o'clock, and the behavior of the recognition object A at 9 o'clock and off work at 18 o'clock can be recorded. Track, automatically completes the commuting record of the recognition object A; if there are multiple behavioral trajectories, the recognition object A appears before the access control at 9:00, 12:00, 13:00, 15:00, 16:00, and 18:00. According to the time, The earliest time is 9:00 to obtain the clock-in record of the identification object A, and the latest time is 18:00 is the clock-out record of the identification object A; Time, you can get statistics on the daily behavior trajectory of the residents of the community entering and leaving the community. Further, according to the daily behavior trajectories of the residents of the community entering and leaving the community, the daily behavior habits of the residents in the community, such as work and rest habits, can be obtained. According to the work and rest habits of the residents of the community (normally, the residents of the community leave the community at the earliest at 5:00, and the residents of the community enter the community at the latest at 23:00). 5:00-23:00), the access control is completely closed during other time periods, and no one can enter or exit through these access control. Only one or two access control is reserved in the community for the residents of the community to conduct access control verification. In this way, the security of the community is improved, and it is also better to manage the population entering and leaving the community, avoiding theft, pyramid schemes, small advertisements and other personnel entering and leaving, causing harassment to the residents of the community, and even endangering the loss of life and property safety.

The access control verification can only include the access control verification when entering, and can also include the entry access verification verification and the exit access verification verification.

Step S307 , collecting images captured by the camera as input data for training samples for face recognition, and manually labeling the images, and using the labeled images as pre-labeled output data.

The images captured by the camera and the labeled images can be used as input data and output data for face recognition training samples in the sample library. The second neural network can be optimized and trained by using the collected images captured by the camera and the labeled images, so that the output result of the second neural network is more accurate.

Step S308, if the access control verification fails, record the image and issue an alarm message.

In order to improve access control security, after the access control verification fails, the image can be recorded as evidence, and an alarm message can be issued, such as community access control verification, strangers who behave strangely before the access control, send images (including images of strangers) with them. ) alarm information to the property monitoring center, and at the same time, the alarm sound is issued to deter strangers and ensure the safety of the community.

According to the access control verification method based on face recognition provided by the present invention, the face recognition result corresponding to the image captured by the camera can be quickly and accurately obtained by using the trained second neural network, and the face recognition of the image captured by the camera can be effectively improved. accuracy, while ensuring the processing efficiency of the second neural network. Further, based on the obtained face recognition results, combined with the personal information of the recognized object, the access control corresponding to the authorization setting information is opened for the recognized object, so that the access control verification is more intelligent, more specific, and more secure, and more convenient services for the recognized object are provided. It can also record the behavior trajectory of the recognized object in combination with the time when the face recognition result is obtained, and automatically realize information such as obtaining the commute clock and other information, and obtaining the daily behavior habits of the recognized object. Access control can also be set according to the daily behavior habits of the identified objects to improve access control security. For the failure of access control verification, the image can be recorded and an alarm message can be issued, which is convenient for leaving evidence and reminding in time to avoid danger. By placing the images captured by the camera and manually labeled images into the sample library, the second neural network can be optimized and trained to make the output results of the second neural network more accurate.

FIG. 4 shows a functional block diagram of an access control verification device based on face recognition according to an embodiment of the present invention. As shown in FIG. 4 , the device includes:

The acquiring module 410 is adapted to acquire an image captured by a camera.

The camera can capture real-time monitoring images, such as installing cameras at the entrance of the community, garage entrance, elevator room, company and other places, it is very convenient to view the real-time situation. The acquisition module 410 acquires an image captured by a camera. In this embodiment, the person in the image is identified. Therefore, the acquisition module 410 acquires an image including a person captured by the camera for subsequent processing of the image.

The recognition module 420 is adapted to input the image into the second neural network to obtain a face recognition result.

The second neural network is a shallow neural network, which has a small number of layers and a fast calculation speed, and is generally suitable for mobile devices, small calculators and other devices. The number of layers of the first neural network is greater than the number of layers of the second neural network. The accuracy of the first neural network is higher. Therefore, the pre-trained output data of at least one intermediate layer of the first neural network is used to guide the training of the second neural network, so that the final output data of the second neural network is the same as that of the first neural network. The final output data of the neural network is consistent, and the computing performance of the second neural network is greatly improved on the premise of retaining the computing speed of the second neural network. The second neural network is obtained by guiding training using the output data of at least one intermediate layer of the pre-trained first neural network, wherein the samples used in the training of the first neural network and the second neural network are both training samples for object recognition.

The recognition module 420 inputs the image into the second neural network to obtain a face recognition result. Wherein, the face recognition result may be the recognition result of the frontal face or the side face of a certain angle of the person in the image.

The judgment module 430 is adapted to use the face recognition result to judge whether the access control verification is successful.

The judgment module 430 utilizes the face recognition result, if the recognition module 420 obtains the face recognition result as the identification object A, the judgment module 430 searches for the existence of A from the user information stored in the access control database, and if so, the judgment module 430 judges that the access control verification is successful, Execute the opening module 440; if not, the judging module 430 judges that the access control verification fails, and does not open the access control to the identified object. Or the recognition module 420 obtains that the face recognition result is the recognition object A, and the judgment module 430 searches for the recognition object A from the user information stored in the access control database, and the information of the recognition object A is on-the-job, then the judgment module 430 judges that the access control verification is successful, and executes the opening module. 440: The information of the identification object A is resignation, then the judgment module 430 judges that the access control verification fails, and does not open the access control for the identification object.

The opening module 440 is adapted to, if yes, complete the access control verification, and open the access control to the identified object.

The opening module 440 completes the access control verification process, and opens the access control for the recognized object identified by the face in the image.

The opening module 440 is further adapted to obtain the personal information of the recognized object according to the face recognition result; according to the personal information of the recognized object, open the access control corresponding to the permission setting information of the recognized object.

After the judging module 430 judges that the access control verification is successful, according to the face recognition result, the opening module 440 further obtains the personal information of the recognized object from the access control database. Among them, the personal information includes permission setting information for the identification object. Specifically, the personal information of the identification object includes that he is a resident on the 5th floor of Building 1 of the community, and the permission setting information is that he can use the elevator of Building 1, but only the elevator buttons on the 5th floor can be used. The judgment module 430 uses the face recognition result to judge that the elevator of Building 1 has successfully authenticated the access control of the identified object, and the opening module 440 opens the elevator door of the elevator of Building 1. The opening module 440 can automatically light up the elevator buttons on the 5th floor according to the acquired personal information of the identification object, especially the authority setting information, so that the identification object can directly reach the fifth floor of Building 1; or, the opening module 440 can When the recognition object manually presses the elevator button on the 5th floor, the elevator button on the 5th floor will be lit. However, when the identification object manually presses the elevator buttons of other floors, the opening module 440 will not light up the elevator buttons of other floors, and the identification object is not allowed to use the elevator buttons of other floors other than the permission setting information, so as to ensure the internal safety of the residents of the community. It is possible to control the possible hidden problems within the residents of the community.

Further, if there are multiple persons in the image acquired by the acquisition module 410, the face recognition results acquired by the recognition module 420 are also multiple. The judgment module 430 uses a plurality of face recognition results to judge whether the access control verification is successful, and the opening module 440 only opens the access control for one or more identification objects that have successfully completed the access control verification. The identification object that fails the access control verification enters trailingly.

According to the access control verification device based on face recognition provided by the present invention, the image captured by the camera is obtained; the image is input into the second neural network to obtain the face recognition result; wherein, the second neural network uses the pre-trained first neural network The output data of at least one middle layer of the first neural network is obtained by guiding the training, and the number of layers of the first neural network is more than the number of layers of the second neural network; the face recognition result is used to judge whether the access control verification is successful; if so, the access control verification is completed, and the Identify the object to open the door. In the present invention, the output data of at least one intermediate layer of the first neural network with a higher number of layers is used to guide the training of the second neural network with a smaller number of layers, so that the second neural network obtained by training keeps its fast calculation condition. , greatly improving its accuracy. The second neural network can be used to quickly and accurately calculate the face recognition result, which is convenient for judging whether the access control verification is successful according to the obtained face recognition result. time, and improve the user experience of the identification object for access control verification.

FIG. 5 shows a functional block diagram of an access control verification device based on face recognition according to another embodiment of the present invention. As shown in FIG. 5 , compared with FIG. 4 , the device further includes:

The recording module 450 is adapted to use the face recognition result to record the behavior track of the recognized object in combination with the time when the face recognition result is obtained.

The recording module 450 uses the face recognition result and the time when the face recognition result is obtained during the recognition, associates the time with the recognition object, and can record the behavior track of the recognition object at each time. The recording module 450 can also automatically acquire the required information according to the behavior track. Such as commuting card information, daily behavior habits, etc. Specifically, for example, the company's access control verification, according to the recognition module 420, the face recognition result of the recognition object A is obtained at 9 o'clock, and the face recognition result of the recognition object A is obtained at 18 o'clock, and the recording module 450 can record the recognition object A at 9 o'clock. The behavior track of going to get off work and leaving get off work at 18:00 will automatically complete the commuting record of the recognition object A; if there are multiple behavior tracks, the recognition module 420 will recognize the object A at 9:00, 12:00, 13:00, 15:00, 16:00, 18 If the point appears before the access control, the recording module 450 can obtain the check-in record for the identification object A at 9:00 at the earliest, and the punch-in record for the identification object A at 18:00 at the latest; or for community access control verification, the recording module 450 uses Multiple face recognition results, and the time when the face recognition results are obtained during the recognition, can count the behavioral trajectories of the residents of the community entering and leaving the community on a daily basis. Further, the recording module 450 can obtain the daily behavior habits, such as work and rest habits, of the residents of the residential area according to the behavior trajectories of the residents of the residential area entering and leaving the residential area on a daily basis. According to the work and rest habits of the residents of the community (normally, the residents of the community leave the community at the earliest at 5:00, and the residents of the community enter the community at the latest at 23:00). 5:00-23:00), the access control is completely closed during other time periods, and no one can enter or exit through these access control. Only one or two access control is reserved in the community for the residents of the community to conduct access control verification. In this way, the security of the community is improved, and it is also better to manage the population entering and leaving the community, avoiding theft, pyramid schemes, small advertisements and other personnel entering and leaving, causing harassment to the residents of the community, and even endangering the loss of life and property safety.

The access control verification can only include the access control verification when entering, and can also include the entry access verification verification and the exit access verification verification.

The alarm module 460 is adapted to record an image and issue an alarm message if the access control verification fails.

In order to improve access control security, after the judgment module 430 judges that the access control verification fails, the alarm module 460 can record the image as evidence, and send out alarm information, such as community access control verification, there is a strange stranger in front of the access control, the judgment module 430 Judging that the access control verification fails, the alarm module 460 sends alarm information carrying images (including images of strangers) to the property monitoring center, and at the same time, emits an alarm sound to deter strangers and ensure community safety.

The face recognition guidance training module 470 includes: a first output unit 471 , a second output unit 472 , a guidance training unit 473 , and may also include a downsampling unit 474 .

The first output unit 471 is adapted to input the training sample data of face recognition into the first neural network obtained by training, and obtain the output data of at least one first intermediate layer of the first neural network.

The first neural network is a pre-trained and solidified neural network. Specifically, the first neural network has been trained using multiple training sample data for face recognition in advance, and the first neural network has been well suited for face recognition. Among them, the first neural network preferably uses a deep neural network, such as a neural network applied to a cloud server, which has good performance, large amount of calculation, high accuracy and slow speed. The first neural network can output the output data of the first intermediate layer of multiple layers. For example, the first neural network contains 4 layers of the first intermediate layer, which are the first intermediate layer of the fourth layer, the first intermediate layer of the third layer, and the first intermediate layer of the second layer. The first intermediate layer and the first intermediate layer are the first intermediate layer, wherein the first intermediate layer of the first layer is the bottleneck layer of the first neural network.

The first output unit 471 inputs the training sample data of face recognition into the first neural network, and can obtain output data of at least one first intermediate layer of the first neural network. Here, the first output unit 471 may only acquire the output data of the first intermediate layer of one layer, or may acquire the output data of the first intermediate layer of adjacent layers, or the first output unit 471 may acquire the first intermediate layer of the multiple layers spaced from each other. The output data of an intermediate layer is specifically set according to the actual situation of the implementation, which is not limited here.

The second output unit 472 is adapted to input the training sample data of face recognition into the second neural network to be trained, and obtain the output data and final output data of at least one second intermediate layer of the second neural network, at least one The second intermediate layer has a corresponding relationship with the at least one first intermediate layer.

The second neural network is the neural network to be trained in the guidance training of the face recognition network, which is a shallow neural network, such as a neural network applied to a mobile terminal, which has limited computing power and poor performance. The first neural network has more layers than the second neural network. For example, the number of layers of the first neural network is 4, which are the first middle layer of the fourth layer, the first middle layer of the third layer, the first middle layer of the second layer and the first middle layer of the first layer; the second neural network The number of layers is 2, which are the second intermediate layer of the second layer and the second intermediate layer of the first layer.

The second output unit 472 inputs the training sample data of face recognition into the second neural network, and obtains output data of at least one second intermediate layer of the second neural network. Wherein, at least one second intermediate layer has a corresponding relationship with at least one first intermediate layer. For example, the first intermediate layer of the first neural network corresponds to the first intermediate layer of the second neural network, and the second intermediate layer of the first neural network corresponds to the second intermediate layer of the second neural network. The two middle layers correspond.

The output data of the second middle layer of the second neural network obtained by the second output unit 472 needs to correspond to the obtained output data of the first middle layer of the first neural network. If the first output unit 471 obtains the output data of the first middle layer of the first neural network For the output data of the two first intermediate layers, the second output unit 472 also needs to obtain the output data of the two second intermediate layers of the second neural network. For example, the first output unit 471 obtains the output data of the first layer of the first neural network and the first middle layer of the second layer, and the corresponding second output unit 472 obtains the first layer of the second neural network and the second middle layer of the second layer. The output data of the layer.

Preferably, at least one layer of the first intermediate layer may include the bottleneck layer of the first neural network, that is, the first layer of the first intermediate layer of the first neural network, and at least one layer of the second intermediate layer may include the bottleneck layer of the second neural network, That is, the first layer and the second intermediate layer of the second neural network. The bottleneck layer is the highest layer of the hidden layer in the neural network, and the middle layer with the least dimension of the output vector. Using the bottleneck layer can ensure that the final output data is more accurate and better training results are obtained when the subsequent guidance training unit 473 performs training.

The second output unit 472 inputs the training sample data of face recognition into the second neural network to be trained, in addition to obtaining the output data of at least one second intermediate layer of the second neural network, the second output unit 472 also The final output data of the second neural network needs to be obtained, so as to use the final output data to calculate the loss and train the second neural network.

The down-sampling unit 474 is adapted to perform down-sampling processing on the training sample data for face recognition, and use the processed data as the training sample data for face recognition of the second neural network.

Considering that the second neural network is a shallow neural network, when the training sample data for face recognition is large, directly using the training sample data for face recognition will affect the operation speed of the second neural network. Optionally, the down-sampling unit 474 can first perform down-sampling processing on the training sample data of face recognition. For example, when the training sample data of face recognition is a picture, the down-sampling unit 474 can perform down-sampling processing to reduce the picture resolution first, The processed face recognition training sample data is used as the face recognition training sample data input by the second neural network. In this way, the second output unit 472 uses the low-resolution face recognition training sample data after downsampling for training, the first output unit 471 uses the high-resolution face recognition training sample data for training, and instructs the training unit 473 to use When the output data of the two neural networks are trained, the second neural network can also obtain high-resolution output results for the low-resolution face recognition training sample data. The guidance training unit 473 is adapted to utilize the loss between the output data of the at least one second intermediate layer and the output data of the at least one first intermediate layer, and the loss between the final output data and the pre-labeled output data, to The second neural network is trained.

The guidance training unit 473 can update the weight parameters of the second neural network according to the loss between the output data of at least one second intermediate layer and the output data of at least one first intermediate layer, so that the second neural network has at least one first intermediate layer. The output data of the second intermediate layer is as close as possible to the output data of at least one first intermediate layer of the first neural network.

At the same time, the guidance training unit 473 can update the weight parameters of the second neural network according to the loss between the final output data of the second neural network and the pre-labeled output data, so that the final output data of the second neural network is as close as possible to the pre-labeled output data. The labeled output data ensures the accuracy of the final output data of the second neural network. By executing the above units, the training of the second neural network is completed. Optionally, when the face recognition guidance training module 470 includes the downsampling unit 474, the downsampling unit 474 also needs to pre-label the training sample data of the downsampling processed face recognition to obtain the downsampling processed face recognition. The pre-labeled output data of the training sample data. The guidance training unit 473 can update the weight parameters of the second neural network according to the loss between the final output data of the second neural network and the pre-labeled output data after the downsampling process, so that the final output data of the second neural network is as large as possible. To be close to the pre-labeled output data of the down-sampling processed data to ensure the accuracy of the final output data of the second neural network.

The collection module 480 is adapted to collect images captured by the camera as input data for training samples of face recognition, and to manually label the images, and use the labeled images as pre-labeled output data.

The collection module 480 collects the images captured by the camera and the marked images, which can be used as input data and output data for training samples in the sample database for face recognition. The images captured by the camera and the labeled images collected by the collection module 480 can be used to optimize the training of the second neural network, so that the output result of the second neural network is more accurate.

According to the access control verification device based on face recognition provided by the present invention, the face recognition result can be obtained quickly and accurately by using the trained second neural network, the accuracy rate of face recognition on the image captured by the camera is effectively improved, and the guarantee of The processing efficiency of the second neural network. Further, the behavioral trajectory of the recognized object can be recorded in combination with the time when the face recognition result is obtained, to automatically realize information such as obtaining commuting and punching cards, and obtaining the daily behavioral habits of the recognized object. Access control can also be set according to the daily behavior habits of the identified objects to improve access control security. For the failure of access control verification, the image can be recorded and an alarm message can be issued, which is convenient for leaving evidence and reminding in time to avoid danger. By placing the images captured by the camera and manually labeled images into the sample library, the second neural network can be optimized and trained to make the output results of the second neural network more accurate.

The present application also provides a non-volatile computer storage medium, where the computer storage medium stores at least one executable instruction, and the computer executable instruction can execute the face recognition-based access control verification method in any of the above method embodiments .

FIG. 6 shows a schematic structural diagram of a computing device according to an embodiment of the present invention. The specific embodiment of the present invention does not limit the specific implementation of the computing device.

As shown in FIG. 6 , the computing device may include: a processor (processor) 602 , a communications interface (Communications Interface) 604 , a memory (memory) 606 , and a communication bus 608 .

in:

The processor 602 , the communication interface 604 , and the memory 606 communicate with each other through the communication bus 608 .

The communication interface 604 is used to communicate with network elements of other devices such as clients or other servers.

The processor 602 is configured to execute the program 610, and specifically may execute the relevant steps in the above embodiments of the face recognition-based access control verification method.

Specifically, the program 610 may include program code including computer operation instructions.

The processor 602 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention. The one or more processors included in the computing device may be the same type of processors, such as one or more CPUs; or may be different types of processors, such as one or more CPUs and one or more ASICs.

The memory 606 is used to store the program 610 . Memory 606 may include high-speed RAM memory, and may also include non-volatile memory, such as at least one disk memory.

The program 610 may be specifically configured to cause the processor 602 to execute the face recognition-based access control verification method in any of the above method embodiments. For the specific implementation of the steps in the program 610, reference may be made to the corresponding descriptions in the corresponding steps and units in the above embodiments of access control verification based on face recognition, which are not repeated here. Those skilled in the art can clearly understand that, for the convenience and brevity of description, for the specific working process of the above-described devices and modules, reference may be made to the corresponding process descriptions in the foregoing method embodiments, which will not be repeated here.

The algorithms and displays provided herein are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used with teaching based on this. The structure required to construct such a system is apparent from the above description. Furthermore, the present invention is not directed to any particular programming language. It is to be understood that various programming languages may be used to implement the inventions described herein, and that the descriptions of specific languages above are intended to disclose the best mode for carrying out the invention.

In the description provided herein, numerous specific details are set forth. It will be understood, however, that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

Similarly, it is to be understood that in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together into a single embodiment, figure, or its description. This disclosure, however, should not be construed as reflecting an intention that the invention as claimed requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, the claims following the Detailed Description are hereby expressly incorporated into this Detailed Description, with each claim standing on its own as a separate embodiment of this invention.

Those skilled in the art will understand that the modules in the device in the embodiment can be adaptively changed and arranged in one or more devices different from the embodiment. The modules or units or components in the embodiments may be combined into one module or unit or component, and further they may be divided into multiple sub-modules or sub-units or sub-assemblies. All features disclosed in this specification (including accompanying claims, abstract and drawings) and any method so disclosed may be employed in any combination, unless at least some of such features and/or procedures or elements are mutually exclusive. All processes or units of equipment are combined. Each feature disclosed in this specification (including accompanying claims, abstract and drawings) may be replaced by alternative features serving the same, equivalent or similar purpose, unless expressly stated otherwise.

Furthermore, those skilled in the art will appreciate that although some of the embodiments described herein include certain features, but not others, included in other embodiments, that combinations of features of different embodiments are intended to be within the scope of the invention within and form different embodiments. For example, in the following claims, any of the claimed embodiments may be used in any combination.

Various component embodiments of the present invention may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that a microprocessor or a digital signal processor (DSP) may be used in practice to implement some or all of some or all of the components in the apparatus for access control verification based on face recognition according to the embodiments of the present invention. Full functionality. The present invention can also be implemented as apparatus or apparatus programs (eg, computer programs and computer program products) for performing part or all of the methods described herein. Such a program implementing the present invention may be stored on a computer-readable medium, or may be in the form of one or more signals. Such signals may be downloaded from Internet sites, or provided on carrier signals, or in any other form.

It should be noted that the above-described embodiments illustrate rather than limit the invention, and that alternative embodiments may be devised by those skilled in the art without departing from the scope of the appended claims. In the claims, any reference signs placed between parentheses shall not be construed as limiting the claim. The word "comprising" does not exclude the presence of elements or steps not listed in a claim. The word "a" or "an" preceding an element does not exclude the presence of a plurality of such elements. The invention can be implemented by means of hardware comprising several different elements and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by one and the same item of hardware. The use of the words first, second, and third, etc. do not denote any order. These words can be interpreted as names.

Claims (22)

1. A door control verification method based on face recognition comprises the following steps:

acquiring an image shot by a camera;

inputting the image into a second neural network to obtain a face recognition result; the second neural network is obtained by utilizing output data of at least one middle layer of a first neural network trained in advance to conduct guiding training, wherein the number of layers of the first neural network is more than that of the second neural network; the training process of the second neural network comprises the following steps: inputting training sample data of face recognition into a first neural network obtained through training, and obtaining output data of at least one first intermediate layer of the first neural network; inputting training sample data of face recognition into a second neural network to be trained to obtain output data and final output data of at least one second intermediate layer of the second neural network, wherein the at least one second intermediate layer and the at least one first intermediate layer have a corresponding relation; training a second neural network by using the loss between the output data of the at least one second intermediate layer and the output data of the at least one first intermediate layer and the loss between the final output data and the pre-labeled output data;

judging whether the access control verification is successful or not by using the face recognition result;

if so, completing the access control verification and opening the access control for the identified object.

2. The method of claim 1, wherein the gate verification comprises an entry verification and/or an exit verification.

3. The method according to claim 1 or 2, wherein the method comprises:

and recording the behavior track of the recognition object by using the face recognition result and combining the time of obtaining the face recognition result.

4. The method of claim 1 or 2, wherein said completing access verification and enabling access to the identified object further comprises:

acquiring personal information of an identification object according to the face identification result; wherein the personal information includes authority setting information for identifying an object;

and opening the access control corresponding to the authority setting information of the identification object according to the personal information of the identification object.

5. The method according to claim 1 or 2, wherein the method further comprises:

and if the access control verification fails, recording the image and sending alarm information.

6. The method of claim 1, wherein the at least one first intermediate layer comprises a bottleneck layer of a first neural network; the at least one second intermediate layer comprises a bottleneck layer of a second neural network.

7. The method of claim 1 or 6, wherein said training a second neural network with a loss between the output data of the at least one second intermediate layer and the output data of the at least one first intermediate layer, and a loss between the final output data and pre-labeled output data further comprises:

and updating the weight parameters of the second neural network according to the loss between the output data of the at least one second intermediate layer and the output data of the at least one first intermediate layer, updating the weight parameters of the second neural network according to the loss between the final output data and the pre-labeled output data, and training the second neural network.

8. The method of claim 1 or 6, wherein before said inputting the input data of the training samples into the second neural network to be trained, obtaining the output data of at least one layer of the second intermediate layer of the second neural network and the final output data, the method further comprises:

and performing downsampling processing on the training sample data of the face recognition, and taking the processed data as training sample data of the face recognition of a second neural network.

9. The method of claim 8, wherein training a second neural network with a loss between the output data of the at least one layer of second intermediate data and the output data of the at least one layer of first intermediate data, and a loss between the final output data and pre-labeled output data further comprises:

and training a second neural network by using the loss between the output data of the at least one second intermediate layer and the output data of the at least one first intermediate layer and the loss between the final output data and the pre-labeled output data of the training sample data for face recognition after the down-sampling processing.

10. The method according to claim 1 or 2, wherein the method further comprises:

collecting images shot by a camera as training sample input data of face recognition, carrying out manual annotation on the images, and using the annotated images as pre-annotated output data.

11. The utility model provides an entrance guard verifying attachment based on face identification, it includes:

the acquisition module is suitable for acquiring images shot by the camera;

the recognition module is suitable for inputting the image into a second neural network to obtain a face recognition result; the second neural network is obtained by utilizing output data of at least one middle layer of a first neural network trained in advance to conduct guiding training, wherein the number of layers of the first neural network is more than that of the second neural network;

the judging module is suitable for judging whether the entrance guard verification is successful or not by utilizing the face recognition result;

the opening module is suitable for completing entrance guard verification and opening an entrance guard for the identified object if the identification is positive;

the face recognition network guidance training module comprises:

the first output unit is suitable for inputting training sample data of face recognition into a first neural network obtained through training to obtain output data of at least one first middle layer of the first neural network;

the second output unit is suitable for inputting training sample data of face recognition into a second neural network to be trained to obtain output data and final output data of at least one layer of second intermediate layer of the second neural network, and the at least one layer of second intermediate layer and the at least one layer of first intermediate layer have a corresponding relation;

and the guiding training unit is suitable for training the second neural network by utilizing the loss between the output data of the at least one second intermediate layer and the output data of the at least one first intermediate layer and the loss between the final output data and the pre-labeled output data.

12. The apparatus of claim 11, wherein the gate verification comprises an entry verification and/or an exit verification.

13. The apparatus of claim 11 or 12, wherein the apparatus further comprises:

and the recording module is suitable for recording the behavior track of the recognition object by using the face recognition result and combining the time of obtaining the face recognition result.

14. The apparatus of claim 11 or 12, wherein the opening module is further adapted to:

acquiring personal information of an identification object according to the face identification result; wherein the personal information includes authority setting information for identifying an object; and opening the access control corresponding to the authority setting information of the identification object according to the personal information of the identification object.

15. The apparatus of claim 11 or 12, wherein the apparatus further comprises:

and the alarm module is suitable for recording the image and sending alarm information if the access control verification fails.

16. The apparatus of claim 11, wherein the at least one first intermediate layer comprises a bottleneck layer of a first neural network; the at least one second intermediate layer comprises a bottleneck layer of a second neural network.

17. The apparatus according to claim 11 or 16, wherein the coaching unit is further adapted to:

and updating the weight parameters of the second neural network according to the loss between the output data of the at least one second intermediate layer and the output data of the at least one first intermediate layer, updating the weight parameters of the second neural network according to the loss between the final output data and the pre-labeled output data, and training the second neural network.

18. The apparatus of claim 11 or 16, wherein the face recognition network guidance training module further comprises:

and the down-sampling unit is suitable for performing down-sampling processing on the training sample data of the face recognition, and taking the processed data as the training sample data of the face recognition of the second neural network.

19. The apparatus of claim 18, wherein the coaching unit is further adapted to:

and training a second neural network by using the loss between the output data of the at least one second intermediate layer and the output data of the at least one first intermediate layer and the loss between the final output data and the pre-labeled output data of the training sample data for face recognition after the down-sampling processing.

20. The apparatus of claim 11 or 12, wherein the apparatus further comprises:

and the collection module is suitable for collecting the image shot by the camera as training sample input data of face recognition, carrying out manual annotation on the image and using the annotated image as pre-annotated output data.

21. A computing device, comprising: the system comprises a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface complete mutual communication through the communication bus;

the memory is used for storing at least one executable instruction, and the executable instruction causes the processor to execute the operation corresponding to the entrance guard verification method based on the face recognition according to any one of claims 1 to 10.

22. A computer storage medium having at least one executable instruction stored therein, the executable instruction causing a processor to perform operations corresponding to the access control verification method based on face recognition according to any one of claims 1 to 10.

Images