CN107967660B - A secure examination system for automatic facial recognition - Google Patents

Abstract

The invention provides an automatic facial recognition security examination system, comprising a monitoring camera and an examination device, both of which are connected to a cloud server in communication; the surveillance camera is used for regularly photographing the examinee's face to obtain the examinee's facial features data, and transmit the facial feature data to the cloud server; the facial feature data is in one-to-one correspondence with the examinee’s information; the cloud server is used to search for the examinee’s identification according to the examinee’s facial data information, and automatically obtain the key words of the examination questions according to the examinee’s identification , thereby constructing test questions for distribution to candidates. The invention solves the problem that the current laboratory safety examination cannot achieve "according to room conditions", individualization and differentiation, and can organize safety examination training in a targeted and more specific manner; Substitute exams, fraud, etc.

Description

A secure examination system for automatic facial recognition

technical field

The invention relates to the field of systems for conducting laboratory safety examinations in universities or scientific research institutes, and more particularly to a safety examination system for automatic facial recognition.

Background technique

Laboratories are important bases for scientific and technological innovation and personnel training in universities or research institutes. However, laboratory safety accidents have occurred frequently, causing many personal and property losses, and seriously endangering the safety and stability of society. Therefore, colleges and universities and scientific research institutes pay more and more attention to laboratory safety education and management. Many domestic colleges and universities offer safety training and guidance, and conduct safety knowledge assessments for freshmen and new faculty members, but the content of safety examinations are often outdated and stereotyped; Single, it takes a lot of manpower and material resources to organize the exam; and some so-called online security exams only need to log in to the account to answer the questions. At present, laboratory safety examinations are generally unable to achieve "according to room conditions", personalized and differentiated educational effects, and at the same time, the monitoring of online examinations is far from enough.

SUMMARY OF THE INVENTION

In order to solve the above technical problems, the present invention provides a security examination system for automatic facial recognition. There are two main purposes of the present invention, one is to solve the problem that the current laboratory safety examination cannot achieve "according to room conditions", individualization and differentiation, and can provide targeted and more specific safety examination training; The security examination is poorly monitored, candidates are prone to substitute examinations, fraud and other problems. Through the identification and real-time monitoring of candidates, remote examinations can be carried out anytime and anywhere, and the strictness of invigilation is improved.

In order to achieve the above object, the present invention provides the following.

A security examination system for automatic facial recognition, the security examination system for automatic facial recognition comprises a surveillance camera and an examination device, the surveillance camera and the examination device are both connected in communication with a cloud server; the surveillance camera is used for regularly photographing candidates face to obtain the candidate's facial feature data, and transmit the facial feature data to the cloud server; the facial feature data is in one-to-one correspondence with the candidate's information;

The cloud server is used to search for the examinee's identification according to the examinee's facial data information, and automatically obtain the key words of the examination questions according to the examinee's identification, so as to construct the examination questions for issuing to the examinee.

Further, the cloud server executes the following candidate identification method:

Step 1: Preprocess the preset photo, locate the facial organs in the preset photo, and construct a feature vector corresponding to the preset photo according to the positioning result, where the feature vector is a known feature vector.

Step 2: Acquire an image captured by a surveillance camera, preprocess the image, locate facial organs, and construct a feature vector to be matched.

Step 3: Perform similarity matching between the known feature vector and the feature vector to be matched.

Step 4: If the matching value is higher than the identification threshold, the matching is successful; otherwise, the matching fails.

Further, the face preprocessing stage includes: image enhancement, binarization, edge detection and image size normalization.

Further, the cloud server utilizes the grayscale difference projection algorithm to locate the facial contour. In the process of locating the facial organs, the eyes are the main organs for locating the nose and the mouth. In the locating process, the integral projection method is used. Combined with the prior knowledge of the face, the feature vector is constructed by using the located geometric position relationship.

Further, the image enhancement includes changing the grayscale histogram of the original image from a certain grayscale interval in the comparison set to a uniform distribution in the entire grayscale range, so as to facilitate the nonlinear adjustment of the image and redistribute the image. Pixel value, so that the number of pixels in a certain grayscale range is roughly the same.

Further, the image enhancement method is a histogram correction method based on cumulative score function transformation.

Further, the size normalization includes scaling the image according to the scaling factor to obtain a calibration image with a uniform size; for the scaling of the image, first, the addition, deletion and movement of pixels are completed according to the size of the desired image, and at the same time, the A grayscale interpolation algorithm needs to be used to keep the image as undistorted as possible.

The beneficial effects of the present invention are as follows: the problem that the current laboratory safety examination cannot be implemented "according to room conditions", individuation and differentiation can be targeted and more specific to organize safety examination training; Due to poor monitoring, candidates are prone to substitute examinations, fraud and other problems, through the identification and real-time monitoring of candidates, remote examinations can be carried out anytime and anywhere, and the strictness of invigilation is improved.

Description of drawings

Fig. 1 is the schematic diagram of the safety examination system of automatic facial recognition provided by the present embodiment 1;

Fig. 2 is the schematic diagram of the acquisition method of examination question keywords provided by the present embodiment 1;

3 is a schematic diagram of a cloud server provided in Embodiment 1;

Fig. 4 is the schematic diagram of the test question allocation module that the present embodiment 1 provides;

Fig. 5 is the schematic diagram of the safety examination method that the present embodiment 1 provides;

Fig. 6 is the schematic diagram of the safety examination system of automatic facial recognition provided by the present embodiment 2;

7 is a schematic diagram of the execution flow of the admission test provided by the present embodiment 2;

8 is a schematic diagram of the execution flow of the final exam provided by the present embodiment 2;

Fig. 9 is the schematic diagram of the execution flow of the examination before receiving the hazardous chemicals/special equipment provided in the present embodiment 2 after purchasing;

Fig. 10 is the schematic diagram of the execution flow of the examination for checking omissions and filling vacancies for problems existing in the inspection after the security inspection provided by the present embodiment 2;

11 is a schematic diagram of the execution flow of the examination of the relevant knowledge that needs to be passed before launching a new project provided by the present embodiment 2;

12 is a schematic diagram of a candidate identification method provided by the present embodiment 3;

Detailed ways

In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings.

Example 1:

The embodiment of the present invention provides a security examination system for automatic facial recognition, and the security examination system for automatic facial recognition consists of hardware equipment and supporting software.

As shown in FIG. 1 , the security examination system for automatic facial recognition includes a plurality of examination rooms 1 , each examination room 1 is equipped with one or more examination stations 2 , and on each examination station 2 Equipped with a monitoring camera 4 and an examination device 3, the surveillance camera 4 is used for regularly photographing the examinee's face to obtain the examinee's facial feature data, and the facial characteristic data is transmitted to the cloud server 5, and the examination device is used to send the examinee to the examinee. Display the test questions, obtain the test taker's answer results, and transmit the answer results to the cloud server.

The surveillance camera 4 and the examination device 3 on each examination station are connected in communication with the cloud server 5 . Each of the test stations has its corresponding ID, and the identification of the surveillance camera and the test device are in a one-to-one correspondence with the ID, so that the cloud server can identify each test worker. bit to manage. Specifically, during the examination process, the cloud server 5 obtains the facial feature data of the examinee at a certain examination station, automatically constructs examination questions, and assigns the examination questions to the examination device 3 to accept the examination. The test questions uploaded by the device 3 are automatically scored for the candidates.

In order to realize the above-mentioned functions, a database is further provided in the embodiment of the present invention, and the database stores the examination question bank and candidate information, and the database is in communication connection with the cloud server 5 .

The design of the database is one of the key contents of the embodiments of the present invention, and is described in detail below:

First, the database records the candidate information, and the recording method is the candidate information database table. The candidate information database table includes six fields: candidate identification, candidate facial data information, candidate professional category, candidate social identity, risk factors to be contacted, and remarks. The candidate ID is the primary key, the candidate's major category can contain one or more professional content, the candidate's social identity indicates the candidate's educational information or employment information, and the note is the candidate's special mark, which will play a role in the automatic construction of examination questions. For example, the candidate's major category is: "Chemistry", "Biology" and/or "Medicine", the candidate's identity: "Master student", "Undergraduate" or "Faculty"; Risk factors to be exposed to: "Sulfuric acid" , "Arsenic trioxide", "Liufu isobutylene" and so on.

In addition, the remarks are used to indicate whether the candidate has entered a subject group, and if the candidate has taken a subject group, the subject group identification is also included in the remarks. If the candidates have already entered the research group, they need to use the research group data management table and the research group database table in the process of automatically constructing the examination questions. The research group management table records the research group identification and the relationship between the research group identification and the research group database table. A research group data management table and a plurality of research group database tables are stored in the database, so that the research group database tables can be quickly located according to the identification of the research group.

The content of the research group database table includes the identification of the research group member, the identity of the research group member, the risk factors that the research group needs to contact, the first-level test-taking information of the research group and the second-level test-taking information of the research group. The status of the research group member is used to represent the candidate's status in the research group, "responsible person" or "ordinary team member", if it is the person in charge, it needs to meet the requirements of the first-level test-taking information of the research group, if it is an ordinary team member , you only need to meet the requirements of the second-level test-taking information of the research group.

The function of the above-mentioned database content in the automatic construction of examination questions is to automatically obtain the keywords of the examination questions according to the information of the candidates. The method for obtaining the keywords of the examination questions is shown in Figure 2, including:

S1. Find the candidate's identity according to the candidate's facial data information.

S2. Find the candidate information database table according to the candidate's identification, and obtain the corresponding candidate's professional category, candidate's social identity, risk factors to be contacted, and remarks.

S3. If the content of the remarks is empty, the candidate's professional category, the candidate's social identity and the risk factors that need to be contacted are used as the key words of the test.

S4. If the content of the remarks is not empty, locate the database table of the subject group corresponding to the candidates according to the remarks.

S5. Obtain the identity of the research group members, the risk factors that the research group needs to contact, the first-level test-taking information of the research group and the second-level test-taking information of the research group in the research group database table.

S6. If the candidate is the person in charge, the candidate's professional category, the candidate's social identity, the risk factors that need to be exposed, the risk factors that the research group needs to be exposed to, and the first-level test-taking information of the research group are used as key words in the test questions.

S7. If the candidate is an ordinary team member, the candidate's professional category, the candidate's social identity, the risk factors that need to be exposed, the risk factors that the research group needs to be exposed to, and the second-level test-taking information of the research group are used as key words in the test questions.

Secondly, the database includes a test question bank, which stores all the test questions that may appear in the test and the corresponding relationship between the test questions and the test question keywords, and can automatically filter out the test question keywords corresponding to the test question keywords according to the test question keywords. All test questions for the cloud server to filter from them to construct test questions for distribution to candidates.

In the question bank, the social identity of the examinee is used to identify the difficulty level of the examinee's examination questions. For example, if the examinee's social identity is an undergraduate student, the selected examination questions are less difficult; if the examinee is a master's student, the selected examination questions are more difficult. The risk factors that need to be exposed, the risk factors that the research group needs to be exposed to, the first-level test-taking information of the research group, and/or the second-level test-taking information of the research group represent the knowledge points that candidates need to test in the exam. For example, if the risk factor to be exposed to contains "arsenic trioxide", the risk factor to be exposed to the research group contains "hexafluoroisobutene", and the level-one test information of the research group includes mastering "sulfuric acid", the knowledge points involved in the test should include " Arsenic trioxide", "hexafluoroisobutene" and/or "sulfuric acid".

On the premise that the database is well constructed, the cloud server realizes a fully automatic and automatic facial recognition security examination system based on the support of the database and the examination station. The cloud server, as shown in Figure 3, includes the following contents:

The integrated control module 51 is used to control the surveillance cameras and examination devices on each examination station.

The candidate identification module 52 is configured to acquire the candidate's facial feature data, and acquire the candidate's information according to the facial feature data.

In order to realize the content of the candidate identification module, it is necessary to pre-enter the database with the facial feature photos of the candidates, as well as the involved professional fields, experimental equipment, experimental consumables and/or possible exposure factors.

Among them, experimental equipment and/or experimental consumables are optional input items. If the experimental equipment and/or experimental consumables are empty, the test keywords will be automatically obtained as shown in Figure 2. If the experimental equipment and/or experimental consumables are not empty, On the basis of the test question keywords obtained in Figure 2, additional experimental equipment and/or experimental consumables are added, and the additional results are used as the final test question keywords; correspondingly, the corresponding experimental equipment and/or experimental consumables are also recorded in the question bank. exam questions.

Further, the cloud server also includes:

The test question allocation module obtains the key words of the test questions of the candidates, constructs the test questions, and distributes the test questions to the test device of the test station where the candidates are located.

The real-time monitoring module is used to regularly photograph candidates' faces to achieve full monitoring during the exam.

The test paper correction module is used for the correction and score calculation of candidates' answer sheets.

The test paper judging and correction module can be used to perform the following steps: step 1, after the test time is over, the system automatically collects the test paper that the examinee has answered; step 2, the system compares the answer of the test paper that has been answered with the standard answer in the question bank. , judge right or wrong; step 3, calculate the final score of the examinee according to the test result.

Specifically, the test question allocation module is shown in Figure 4, including:

The information extraction module 61 is used to obtain the keywords of the test questions.

The degree judgment module 62 is used for judging the difficulty of the test question according to the social identity of the examinee in the keyword of the test question.

The question acquisition and test paper generation module 63 selects the corresponding test questions from the question bank according to the difficulty judgment result of the test questions and the test question keywords, and extracts a part to generate the test paper.

Specifically, there can be one or more question banks in the database, and the question acquisition and test paper generation module can extract a certain number of test questions from different question banks, and form these questions into test papers in random order.

Specifically, the safety examination system of the automatic facial recognition performs the safety examination method as shown in Figure 5 during the working process:

Step 1: Before the exam, the facial images of the candidates are collected through the surveillance cameras on the exam workstations.

In step 2, it is automatically compared with the preset photos. When the comparison does not match, the system prompts a comparison error and the test is terminated; when the comparison matches, the examinee begins to enter the test.

Specifically, there may be only one preset photo, or there may be multiple photos.

Step 3: During the examination, remote real-time monitoring is performed on the candidates, and the candidates are photographed and compared at specified time intervals.

Remote real-time monitoring of candidates can be carried out through surveillance cameras, and candidates can be photographed and compared at specified intervals, such as 2-3 seconds, to prevent candidates from cheating or taking exams in the middle.

In order to more clearly illustrate the specific implementation process of the embodiment of the present invention, an example is as follows:

Candidate Li Ming, pre-entered his personal information in the database: chemistry major, master student, frequent exposure risk factors: potassium cyanide, hydrofluoric acid. And upload the personal one-inch bareheaded electronic photo to the database. He applied for the security exam to start at 9:00 on September 11, 2017. At 8:50 on the same day, a photo of his facial features was taken by a surveillance camera and transmitted to the cloud server through the network. After the computer did a comparison, it was found that the comparison matched, and the examination began. According to his personal information, the cloud server first extracts the keywords: "Chemistry", "Master's degree", "Potassium cyanide", "Hydrofluoric acid", and judges that the level of safety examination suitable for Li Ming is: Chemistry, medium difficulty, " Potassium cyanide" and "hydrofluoric acid" knowledge point topics. Therefore, the system selects appropriate questions from the question bank according to these conditions, generates test papers, and sends them to Li Ming's test terminal. During the exam, the camera takes a photo of facial features every 2-3 seconds and transmits it to the main server. The person in charge of the exam can monitor the exam situation in real time through the main server. After the test, the system automatically collects Li Ming's answer results, corrects the test paper and calculates the score.

Candidate Liu Min, pre-entered his personal information: biology major, faculty member/group leader, laboratory animal research group, frequently exposed risk factors: pathogenic microorganisms, parasites. And upload the personal one-inch bareheaded electronic photo to the cloud server. She applied to take the safety exam starting at 12:00 on October 10, 2017. At 11:50 on the same day, she entered the examination system, and the surveillance camera took a photo of his facial features and transmitted it to the cloud server through the network. After the computer performed the comparison, it found a match and began the examination. Based on her personal information, the system first extracts keywords: "biology", "experimental animal research group" (further, it can be related information of the research group), "faculty and staff", "pathogenic microorganisms", "parasites", and judges that it is suitable for Liu. The safety examination level of Min is: biology, laboratory animal-related, higher difficulty, "pathogenic microorganisms" and "parasites" knowledge points. Therefore, the system selects appropriate questions from the question bank according to these conditions, generates test papers, and sends them to Liu Min's computer terminal for the test. During the exam, the camera will take photos of facial features every 2-3 seconds and transmit them to the cloud server. After the test, the system automatically collects Liu Min's answer results, corrects the test paper and calculates the score.

Embodiment 2:

As shown in FIG. 6 , the embodiment of the present invention uses the automatic facial recognition security examination system shown in Embodiment 1, and specifically describes the usage of the automatic facial recognition security examination system.

The security examination system of automatic facial recognition in the embodiment of the present invention can be used to execute two examination modes: regular examination 7 and dynamic examination 8. The regular examination 7 is a unified and centralized examination mode, and the dynamic examination 8 is a kind of examination mode. Customized, remote exam mode. The regular exams include entrance exams 71 and final exams 72 for related majors. The dynamic test 8 includes the test 81 for hazardous chemicals/special equipment after purchase and before use, the test 82 for checking and filling vacancies after the safety inspection, and the test 83 for related knowledge that needs to be passed before launching a new project.

In the regular test, the test takes the form of a unified location and a centralized test. The identity information of the candidates is checked before the test, and the test papers are automatically distributed according to the information of the candidates. Candidates of the same major draw the same question bank, but the specific questions are not necessarily the same, which can effectively avoid cheating.

The execution process of the admission test is shown in Figure 7, including:

A1: Register an account and fill in personal information.

The account number corresponds to the candidate ID in the database, and personal information also needs to be entered into the database for subsequent automatic questioning.

A2: According to the laboratory and subject group where the candidate is located, the system automatically selects the practice questions of the corresponding major and difficulty, and the candidate conducts self-study and mock exams.

A3: The school arranges the admission test at the same time and place, and the candidates take the online test. The examination papers are distributed by the examination system according to the candidates' information.

A4: After the exam is completed, the system will automatically score and count the number of pass/failures.

The execution flow of the final exam is shown in Figure 8, including:

B1: Safety courses are set for specific majors. As a compulsory or elective course, 0.5 or 1 credit can be set.

B2: The school arranges the admission test at the same time and place, and the candidates take the online test. The examination papers are distributed by the examination system according to the candidates' information.

B3: After the exam is completed, the system will automatically score and count the number of pass/failures. Students who pass the exam receive credits.

The dynamic test adopts the form of separate customization and remote monitoring, which is only for the partial needs of some candidates, and the test papers are automatically distributed according to the candidates' information. Invigilators can perform facial recognition and identity authentication on candidates through online monitoring to complete remote exams. It reduces the burden of organizing a large examination room and a large number of invigilators, and the examination time is flexible, so that safe examinations can be conducted anytime, anywhere.

Among them, the execution process of the examination before receiving the hazardous chemicals/special equipment after purchasing is shown in Figure 9, including:

C1. To purchase hazardous chemicals or special equipment, first make purchases on the purchase mall, and send the items to a special warehouse for temporary storage.

C2. The purchaser should fill in the application for use online, and the examination system will allocate special questions about the safety knowledge related to the hazardous chemicals or special equipment. The purchaser must pass the examination before obtaining the qualification.

C3. All users involved in the hazardous chemicals or special equipment should take the special examination. If it is found later that any person who fails the examination uses the hazardous chemicals or special equipment by himself, he will be ordered to take the examination immediately and conduct criticism and education.

Figure 10 shows the execution flow of the examination to check for omissions and fill vacancies after the security inspection, including:

D1. After each safety inspection, summarize the respective problems and shortcomings of each laboratory and publish them on the relevant website.

D2. Relevant staff in the announced laboratory should apply for the special exam for omissions and filling vacancies before the next inspection, and the exam system will allocate relevant questions that have problems in this safety inspection.

D3. All relevant personnel in each laboratory with problems have passed the examination of filling vacancies, and the safety inspection is not over until this time. If it fails or fails to apply, the next safety inspection will directly determine that it is unqualified, and continue to urge its rectification.

Figure 11 shows the execution flow of the relevant knowledge exams that need to be passed before starting a new project, including:

E1. Under the premise that a research group has applied for a new subject or project, if new safety knowledge is involved, it must be reported to the school safety department before launching a new project.

E2. The school safety management department conducts safety assessment and analysis on the new subject to be carried out, and re-determines the safety examination level that members of the subject group need to pass.

E3. If the re-assessed exam level is higher than the previous one, the experimenters of the subject group are required to retake the safety exam, and the questions will be automatically allocated by the exam system.

Further, candidates can also re-do simulation exercises before the test.

Example 3:

In the above two embodiments, it is necessary to monitor candidates or identify candidates’ information based on facial recognition. In order to improve the recognition rate and improve the recognition speed, the embodiment of the present invention provides a method for identifying candidates. The method is shown in FIG. 12 . shown, including:

Step 1: Preprocess the preset photo, locate the facial organs in the preset photo, and construct a feature vector corresponding to the preset photo according to the positioning result, where the feature vector is a known feature vector.

Step 2: Acquire an image captured by a surveillance camera, preprocess the image, locate facial organs, and construct a feature vector to be matched.

Step 3: Perform similarity matching between the known feature vector and the feature vector to be matched.

Step 4: If the matching value is higher than the identification threshold, the matching is successful; otherwise, the matching fails.

The face preprocessing stage includes: face image enhancement, binarization, edge detection, and image size normalization.

The facial organ localization module determines the facial contour from the original image. In the embodiment of the present invention, the grayscale difference projection algorithm is used to locate the facial contour, and the accumulated value of grayscale difference is used to determine the boundary lines on both sides of the face. The calculation amount is small, and the positioning speed is small. Fast and accurate. In the process of positioning facial organs, the eyes are the main organs for positioning, and its positioning accuracy affects the subsequent organ positioning. Next, the nose tip and mouth are located using the integral projection method combined with facial prior knowledge. Construct the appropriate feature vector using the located geometric position relationship.

Specifically, the image enhancement method in the embodiment of the present invention mainly includes:

The grayscale histogram of the original image is changed from a certain grayscale interval in the comparison set to a uniform distribution in the entire grayscale range, so as to facilitate the nonlinear adjustment of the image, redistribute the image pixel values, and make a certain grayscale range. The number of pixels inside is about the same. Specifically, the image enhancement method is a histogram correction method based on cumulative score function transformation. If the original grayscale of the pixel is R, the transformed grayscale is S, and S is the normalized grayscale value, and its grayscale transformation function T(R) is:

其中P_R是灰度值概率，n_j是图像中第j级灰度的像素总数，n是像素总数，R_j是灰度值像素数目。

where P _R is the gray value probability, n _j is the total number of pixels of the jth gray level in the image, n is the total number of pixels, and R _j is the number of gray value pixels.

Before feature extraction, image preprocessing is usually performed. For face recognition, the first task is to segment the face image and locate the main organs, and the scale of the face image should be normalized. By scaling the image according to the scaling factor, a calibration image with a uniform size is obtained. For image scaling, first of all, an algorithm is needed to define the spatial transformation itself, that is, to complete the addition, deletion and movement of pixels according to the size of the desired image. At the same time, a gray-level interpolation algorithm is needed to keep the image as undistorted as possible.

After normalization, the positions of the eyes, nose tip, mouth and other organs are further determined through the projection curve. This method is fast and simple, and can satisfy weak real-time applications.

During the organ identification process, the eye is located first. The gray value of the eye part in the facial image is usually smaller than the gray value of the surrounding area. Using this feature, the method of integral projection is often used to locate the eyes. The overall gray value of the image in the horizontal or vertical direction can be reflected by the integral projection function, so the position of the pupil can be judged by integral projection of the eye area. In order to further increase the accuracy of eye positioning, on the basis of the integral projection method, the consideration of the horizontal gray level is fused, that is, the gray level of the eye changes greatly in the horizontal direction. Differentiating at the sudden change of gray level will generate a high value, and the absolute value of which will be accumulated, and the line with the larger gray level change will have a larger accumulated value. This will position the eye.

On the basis of the accurate positioning of the eyes, various other organs are subsequently positioned, and then facial recognition can be performed.

The above disclosures are only preferred embodiments of the present invention, and of course, the scope of the rights of the present invention cannot be limited by this. Therefore, equivalent changes made according to the claims of the present invention are still within the scope of the present invention.

Claims (6)

1. a security examination system of automatic facial recognition, is characterized in that, the security examination system of described automatic facial recognition comprises surveillance camera and examination device, and described surveillance camera and examination device are all connected with a cloud server communication; Described monitoring The camera is used to regularly photograph the examinee's face to obtain the examinee's facial feature data, and transmit the facial feature data to the cloud server; the facial feature data is in one-to-one correspondence with the examinee's information; the cloud server is used for searching according to the examinee's facial data information Candidate identification, and automatically obtain the test key words according to the candidate identification, so as to construct the test questions for distribution to the candidates; The cloud server executes the following candidate identification methods: Step 1: preprocessing the preset photo, locating the facial organs in the preset photo, and constructing a feature vector corresponding to the preset photo according to the positioning result, where the feature vector is a known feature vector; Step 2: acquiring the image captured by the surveillance camera, preprocessing the image, locating the facial organs, and constructing the feature vector to be matched; Step 3: Perform similarity matching between the known feature vector and the feature vector to be matched; Step 4: If the matching value is higher than the identification threshold, the matching is successful; otherwise, the matching fails; Also provided with a database, the database stores the examination question bank and candidate information, the database is connected to the cloud server in communication; the database records the candidate information, and the recording method is the candidate information database table, the candidate information database The table includes six fields: candidate ID, candidate facial data information, candidate professional category, candidate social identity, risk factors to be contacted, and remarks; the candidate ID is the primary key, the candidate professional category contains one or more professional content, the candidate social identity Indicates the candidate's academic information or employment information; the remarks are used to indicate whether the candidate has entered the subject group, and if the candidate has already entered a subject group, the remarks also include the subject group identification; if the candidate has entered the subject group, it will be automatically constructed. For the examination questions, it is also necessary to use the research group data management table and the research group database table; the research group data management table records the research group identification and the relationship between the research group identification and the research group database table; the database stores a Zhang research group data management table and multiple research group database tables, so as to quickly locate the research group database table according to the research group identification; the content of the research group database table includes the identification of the research group member, the identity of the research group member, the research group Risk factors that need to be contacted, first-level test-taking information of the research group, and second-level test-taking information of the research group; the status of the research group member is used to indicate the candidate's status in the research group, "the person in charge" or "ordinary team member", if it is The person in charge needs to meet the requirements of the first-level test-taking information of the research group, and if he is an ordinary team member, he only needs to meet the requirements of the second-level test-taking information of the research group; The cloud server includes: an integrated control module for controlling surveillance cameras and examination devices on each examination station; The candidate identification module is used to obtain the candidate's facial feature data, and obtain the candidate's information according to the facial feature data; The test question allocation module obtains the key words of the test questions of the candidates, constructs the test questions, and assigns the test questions to the test device of the test station where the candidates are located; The real-time monitoring module is used to regularly photograph candidates' faces to achieve full monitoring during the exam; The test paper correction module is used for the correction and score calculation of candidates' answer sheets; The method for obtaining the key words of the test questions includes: searching for the examinee's identification according to the examinee's facial data information; Search the candidate information database table according to the candidate identification, and obtain the corresponding candidate's professional category, candidate's social identity, risk factors to be contacted, and remarks; If the content of the remarks is empty, the key words of the test questions will be the candidate's professional category, the candidate's social identity and the risk factors to be exposed to; If the content of the remarks is not empty, locate the database table of the subject group corresponding to the candidates according to the remarks; In the database table of the research group, the identity of the research group member, the risk factors that the research group needs to contact, the first-level test-taking information of the research group and the second-level test-taking information of the research group are obtained; Social identity, risk factors to be contacted, risk factors to be contacted by the research group, and first-level test-taking information of the research group are used as key words in the test questions; , the risk factors that the research group needs to be exposed to and the secondary examination information of the research group as the keywords of the test questions; The monitoring process requires candidates to identify based on facial feature positioning, use the grayscale difference projection algorithm to locate the facial contour, and use the grayscale difference accumulated value to determine the boundary lines on both sides of the face. In the process of locating facial organs, the eyes are the location The main organs of the original image are extracted by image enhancement, and the grayscale histogram of the original image is changed from a certain grayscale interval to a uniform distribution in the entire grayscale range, so as to facilitate the nonlinear extraction of the image. Extend, redistribute the image pixel value, so that the number of pixels in the preset grayscale range is roughly the same, if the original grayscale of the pixel is R, the transformed grayscale is S, and S is the normalized grayscale value, which The grayscale transformation function T(R) is:

, where PR is the gray value probability, nj is the total number of pixels of the jth gray level in the image, n is the total number of pixels, and Rj is the number of gray value pixels; in the process of organ recognition, the eye is first located, and the integral projection function It reflects the overall gray value of the image in the horizontal or vertical direction. On the basis of the integral projection method, differentiation is performed at the sudden change of gray level, which will generate a high value. If the absolute value is accumulated, the greater the gray level change. In that row, the larger the accumulated value, the more the eye is positioned. 2. The security examination system of automatic face recognition according to claim 1, characterized in that: the face preprocessing stage includes: image enhancement, binarization processing, edge detection and image size normalization. 3. the safety examination system of automatic facial recognition according to claim 2, is characterized in that, described cloud server utilizes grayscale difference projection algorithm to locate facial contour, in the process of carrying out the positioning of facial organ, eye is positioned Next, the main organs locate the tip of the nose and the mouth. In the process of localization, the integral projection method is used combined with the prior knowledge of the face, and the feature vector is constructed by using the geometric position relationship obtained by the localization. 4. the security examination system of automatic facial recognition according to claim 3, is characterized in that, described image enhancement comprises the grayscale histogram of original image from a certain grayscale interval in comparison set to become in all grayscale range In order to stretch the image non-linearly, redistribute the image pixel values, so that the number of pixels in a certain gray range is roughly the same. 5 . The security examination system of automatic face recognition according to claim 4 , wherein the method of image enhancement is a histogram correction method based on cumulative distribution function transformation. 6 . 6. the safety examination system of automatic facial recognition according to claim 3, is characterized in that, described size normalization comprises carrying out the scaling of image according to scaling factor, obtains the calibration image with uniform size; For the scaling of image, first , to complete the addition, deletion and movement of pixels according to the size of the desired image, and at the same time, a gray-level interpolation algorithm is required to keep the image as undistorted as possible.

Images