HK1237490B - An anti-cheating network research method, device and system based on bioassay - Google Patents

Description

A method, device and system for anti-cheating network research based on liveness detection

Technical Field

The present invention relates to the field of image recognition technology, and in particular to a method, device and system for preventing cheating in online research based on liveness detection.

Background Art

Currently, with the development of the internet, online surveys have become one of the primary methods for obtaining data for market research. Verifying the authenticity and validity of users during the survey process is a primary issue in determining the validity of questionnaire data samples obtained through online surveys. Existing online questionnaire survey systems primarily perform validity verification during user registration, such as issuing a verification code for user submission and asking users multiple questions based on the validity of their answers. However, since computers are currently very mature in simulating human verification code recognition and submission technology, and questionnaires are often answered by machines instead of humans, the authenticity and validity of online questionnaire survey sample data has been significantly reduced.

Summary of the Invention

In order to overcome the deficiencies of the prior art, one of the objectives of the present invention is to provide an anti-cheating network research method based on liveness detection, which can verify the authenticity of users.

A second object of the present invention is to provide an anti-cheating network research device based on liveness detection, which can verify the authenticity of the user.

A third object of the present invention is to provide an anti-cheating network survey system based on liveness detection, which can verify the authenticity of users.

One of the purposes of the present invention is achieved by the following technical solution:

A method for preventing cheating in online research based on liveness detection includes the following steps:

Model building steps: Establish an action recognition model library;

Information acquisition step: acquiring the user's action recognition information, wherein the action recognition information includes a current feature vector of the person's face;

Feature comparison step: Compare the user's action recognition information with the verification feature vector in the action recognition model library. If the comparison results are consistent, the verification is passed.

Furthermore, the model building step specifically includes the following sub-steps:

Action acquisition step: acquiring verification action information, wherein the verification action information includes a verification feature vector of the human face, and the verification feature vector is a displacement change of a verification feature point;

Steps for establishing an action model library: establish an action model library based on the verification action information and the corresponding operation instructions.

Furthermore, the model building step also includes a facial recognition step: building a facial recognition model library based on the acquired facial recognition information of the user.

Furthermore, the feature comparison step specifically includes the following sub-steps:

Similarity judgment step: judging whether the similarity between the action recognition information and the verification action information in the action recognition model library is greater than a preset value; if so, the verification is passed.

Furthermore, the feature comparison step also includes a face comparison step: comparing the acquired face recognition information with the data in the face recognition model library, and if the comparison results are consistent, executing the similarity judgment step.

The second object of the present invention is achieved by adopting the following technical solution:

A liveness detection-based anti-cheating network research device includes the following modules:

Model building module: used to build action recognition model library;

Information acquisition module: used to obtain the user's action recognition information, the action recognition information includes the current feature vector of the person's face;

Feature comparison module: used to compare the user's action recognition information with the verification feature vector in the action recognition model library. If the comparison results are consistent, the verification is passed.

Furthermore, the model building module specifically includes the following submodules:

Action acquisition module: used to obtain verification action information, the verification action information includes a verification feature vector of a person's face, and the verification feature vector is a displacement change of a verification feature point;

Action model library establishment module: used to establish an action model library based on verification action information and corresponding operation instructions.

Furthermore, the model building module also includes a facial recognition module: used to build a facial recognition model library based on the acquired facial recognition information of the user.

Furthermore, the feature comparison module specifically includes the following submodules:

Similarity judgment module: used to judge whether the similarity between the action recognition information and the verification action information in the action recognition model library is greater than a preset value. If so, the verification is passed.

The third object of the present invention is achieved by adopting the following technical solution:

A liveness detection-based anti-cheating network research system includes an executor configured to execute any one of the liveness detection-based anti-cheating network research methods described above.

Compared with the prior art, the present invention has the following beneficial effects:

The present invention introduces face recognition for liveness detection in an online questionnaire survey system. By adding a user verification link that requires users to complete facial movements according to prompts in combination with liveness detection technology, the validity and authenticity of questionnaire sample data are improved, and a large number of invalid questionnaires that use machines to answer questions fraudulently are avoided.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a flow chart of the anti-cheating network research method based on liveness detection of the present invention;

FIG2 is a structural diagram of the anti-cheating network research device based on liveness detection of the present invention.

DETAILED DESCRIPTION

The present invention will be further described below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments.

The anti-cheating network research system based on liveness detection of the present invention mainly includes: an intelligent device, a camera, and a server.

Smart device: A computer with a camera or a mobile terminal with a camera, such as a mobile phone. Users access the online questionnaire through the smart device and perform related operations such as registration, login, questionnaire settings, and answering questions.

Camera: used to obtain facial video images of users when using the questionnaire system.

The server is equipped with: user management module, questionnaire module, user verification module; the server is connected to the smart device via wireless network or optical cable;

User management module: obtains and manages user data and assigns permissions. It includes registration, login, and user permission management.

Registration: Through the registration process, the user is guided to submit basic identity information, set a password, and is prompted to perform specified actions through the camera to obtain the registered user's facial video image data. The above information is sent to the user rights management module, and the facial recognition model and basic information of each user are established and stored accordingly.

Login: Through the login process, verify the user's identity information, match the user's basic information data, and perform user verification when necessary. After the user logs in successfully, the user information is sent to the user rights management module to determine the user's rights;

User rights management: Storing and managing users' basic profile information and corresponding facial recognition model information, as well as questionnaire setting management permissions or questionnaire answering permissions; configuring users' corresponding questionnaire settings or answering permissions based on the profile information submitted by users during registration and the account type information selected, and performing permission determination and allocation after the user logs in; establishing a facial model corresponding to the user using the facial video obtained during user registration to verify user consistency;

Questionnaire module: includes three parts: questionnaire setting, online questionnaire, and questionnaire data analysis; Questionnaire setting: Questionnaire management users configure questionnaire content, survey questions, and matching user types through the questionnaire setting module, and complete the settings to publish the questionnaire.

Online questionnaires: Users view the questionnaire content, perform corresponding actions to answer the questions, and submit their information. Online questionnaires include survey questions set by the questionnaire administrator and randomly inserted user verification questions. Randomly inserting user verification questions can effectively improve the authenticity of questionnaire data. This involves randomly extracting a facial action instruction configured by the recognition and verification module during the user's response. A video of the user performing the specified facial action is captured through a camera. The video is then compared against the user's facial model to verify consistency and against the recognition model to verify authenticity.

Questionnaire data: After the questionnaire data analysis module obtains the answer information submitted by the user, it analyzes and processes the answer information and displays the questionnaire data results to the questionnaire management user for review.

User verification module: includes user consistency verification and user authenticity verification.

User consistency verification: We extract key frames from the user's facial video images during questionnaire surveys or login verification, perform facial feature extraction on the user, and compare them with the user's facial feature model built based on the facial video images submitted during user registration to verify user consistency. When the similarity is greater than 80%, we consider it to be the same user.

The main principle process is:

Analyze the facial video images submitted by users during registration, extract key frames, and construct 72 key points of each facial part based on the shape, size, position, and distance of facial features such as the iris, nose, and mouth corners. Then, calculate their geometric features, and use these features to form a feature vector describing the face. Build a facial feature vector set for each registered user as the user's facial feature model, which is stored in the user rights management module for model comparison during subsequent user consistency verification.

When a user logs in or performs user verification during the question-answering process, the user's facial video key frames of the verification process are extracted, and the feature vectors of 72 facial points are compared with the facial feature model of the corresponding user to determine the user's consistency.

User authenticity verification includes: recognition model and authenticity verification. The established recognition model is used to analyze the user's facial movement video captured by the camera, extracting the feature vectors of the user's facial movement changes, and comparing them with the recognition model to verify the user's authenticity.

Recognition model: By extracting key frames from facial video images, constructing facial key points, and extracting features from the key points; by learning from a large number of users' facial action videos, a training set template library is established based on the corresponding action instruction information of the key points when the user's facial movements change, which serves as a recognition model for user authenticity verification.

As shown in FIG1 , the present invention provides an anti-cheating network research method based on liveness detection, comprising the following steps:

S1: Establishing an action recognition model library; the model establishment step specifically includes the following sub-steps:

S11: Build a facial recognition model library based on the acquired user's facial recognition information; a person's facial structure and facial features have significant changes when different facial movements change. Through learning and continuous correction, based on the shape, size, position, combination distance and other attributes of the facial features such as the iris, nose, mouth corners, cheekbones, etc., 72 key points that can reflect the changes in facial movements and are stable under the influence of various light projection external environments and when the face is offset at various angles are found. Based on these 72 key points, a recognition model library is established; this step mainly collects the main facial information of the person for later facial recognition verification;

S12: Acquire verification action information, including verification feature vectors of the human face, which are displacement changes of verification feature points; classify the user verification action into five instructions: nodding, turning the head left, turning the head right, blinking, and opening the mouth; and establish a recognition model based on the coordinate offset vectors of 72 facial points corresponding to each action instruction;

S13: Establish an action model library based on the verification action information and the corresponding operation instructions. Through machine learning training and analysis of a large number of user facial action change videos, statistics are collected on the point coordinate information change data of 72 key points when different facial actions change, and their coordinate offset vectors under different action instructions are calculated to form the facial feature vectors of each action instruction; the extracted facial verification feature vectors are stored in the corresponding action instruction template library to establish a recognition model for user verification. The training process requires correcting the vector set of each instruction by continuously comparing the recognition results; this step mainly uses verification actions to identify whether it is a real person answering;

S2: Obtaining the user's action recognition information, which includes the current feature vector of the person's face; this step is mainly used to collect data and can be interspersed with the user's login period or the user's questionnaire answering process;

S3: Compare the user's action recognition information with the verification feature vector in the action recognition model library. If the comparison results are consistent, the verification is passed. The feature comparison step specifically includes the following sub-steps:

S31: Compare the acquired facial recognition information with the data in the facial recognition model library. If the comparison results are consistent, perform a similarity determination step.

S32: Determine whether the similarity between the action recognition information and the verification action information in the action recognition model library is greater than a preset value. If so, the verification is passed; to verify the acquired information.

The main application process of the present invention is:

When a user accesses the online questionnaire survey system, he/she clicks on "Register", submits his/her identity information and account type (normal answering user, questionnaire management user) to start creating a user account;

The user verification module randomly generates a set of liveness detection user verification instructions and uses the camera to obtain facial video images of the user completing the specified facial movements according to the prompts;

Extract the feature vector of the user's facial action in the video image and compare it with the feature vector set of the corresponding action in the recognition model. If the similarity reaches more than 80%, it passes the verification;

The user's facial feature vector is extracted and stored, a user facial feature model is built, and the corresponding registration-related identity information submitted by the user is stored in the user management module. The above steps complete the user registration.

After registration, users will begin the login process when they need to use the questionnaire survey system. Normal account logins require only verification of the user's account and password. If an account has experienced any anomalies (e.g., successfully passing password verification after multiple incorrect password entries), the user verification process will begin.

The user is prompted to perform relevant facial movements, such as opening their mouth or blinking, through the camera. After capturing the user's facial video, the user verification module extracts the user's facial feature vector from the video and compares it with the corresponding user's facial feature model stored in the system to determine the user's consistency. If the consistency is confirmed, the module then performs a specific action consistency check to verify the user's authenticity. If all verifications pass, the user is successfully logged in and can enter the questionnaire survey system to perform relevant operations.

When users enter the questionnaire system and answer questions, questions about specific facial movements are randomly inserted to enhance the authenticity of the questionnaire sample. Specifically, after completing the general questions, the facial movement questions will be entered. The system uses a camera to capture a video of the user performing the specified facial movements as prompted. It then extracts the user's facial feature vectors from the video, compares them to the user's facial feature model for user consistency, and then compares them to the corresponding action feature vector set in the recognition model for user authenticity verification. If verification passes, the user completes the question and proceeds to the next step.

After obtaining the complete answer information submitted by the user, the questionnaire data analysis module analyzes and processes the answer information and displays the questionnaire data results to the questionnaire management user for review.

The above method is the most preferred solution of this embodiment, and it can also be verified by other methods. For example, when performing user consistency judgment, it is used to verify whether the user is the user himself when submitting the questionnaire or other related operations. This step can improve the accuracy of matching the basic information of the user of the questionnaire sample when obtaining the sample data of the specified population, and can prevent other users who do not meet the characteristics from answering questions on behalf of the registered user. This step does not require judgment, and the basic information of the user can be verified by other questioning methods. In the user authenticity verification link, some true or false questions can also be randomly asked to nod for yes and shake their head for no, and the correctness of the answer and the consistency of the feature vector of the user's facial action video image with the recognition model can be judged to complete the real user verification.

As shown in FIG2 , the present invention provides an anti-cheating network research device based on liveness detection, which includes the following modules:

Model building module: used to build an action recognition model library; the model building module specifically includes the following submodules:

Facial recognition module: used to build a facial recognition model library based on the acquired user's facial recognition information;

Action acquisition module: used to obtain verification action information, the verification action information includes a verification feature vector of a person's face, and the verification feature vector is a displacement change of a verification feature point;

Action model library establishment module: used to establish an action model library based on verification action information and corresponding operation instructions.

Information acquisition module: used to obtain the user's action recognition information, the action recognition information includes the current feature vector of the person's face;

Feature comparison module: This module is used to compare the user's motion recognition information with the verification feature vectors in the motion recognition model library. If the comparison results are consistent, the verification is passed. The feature comparison module specifically includes the following submodules:

Facial comparison module: used to compare the acquired facial recognition information with the data in the facial recognition model library. If the comparison results are consistent, the similarity judgment module is executed;

Similarity judgment module: used to judge whether the similarity between the action recognition information and the verification action information in the action recognition model library is greater than a preset value. If so, the verification is passed.

The above embodiments are only preferred embodiments of the present invention and cannot be used to limit the scope of protection of the present invention. Any non-substantial changes and replacements made by technicians in this field on the basis of the present invention fall within the scope of protection required by the present invention.

Claims (10)

1. A method for preventing cheating in online surveys based on liveness detection, characterized by comprising the following steps: Questionnaire setup steps: Questionnaire management users configure questionnaire content, survey question types, and matching user types through the questionnaire settings module, and then publish the questionnaire after setup. The online questionnaire allows users to view the questions, answer them, and submit information. The online questionnaire includes survey questions set by the questionnaire management user and randomly inserted user verification questions, or randomly selected true/false questions requiring users to nod for yes and shake their heads for no. User verification questions are randomly inserted questions involving specific facial expressions. Prompt Steps: Add a user verification step that combines liveness detection technology with prompts to complete facial movements, allowing users to perform specified facial movements according to the prompts; Model building steps: Establish an action recognition model library; Information acquisition steps: Acquire the user's action recognition information, which includes the current feature vector of the human face; the system acquires video images of the user's specified facial actions performed as prompted by the camera, and extracts the user's facial feature vector from the video; Feature comparison step: The user's action recognition information is compared with the verification feature vector in the action recognition model library. That is, the user's facial feature vector extracted from the video is compared with the corresponding action feature vector set in the recognition model. If the comparison results are consistent, the verification is passed. 2. The anti-cheating network survey method based on liveness detection as described in claim 1, characterized in that the model building step specifically includes the following sub-steps: Action acquisition steps: Acquire verification action information, which includes verification feature vectors of the human face, and the verification feature vectors are displacement changes of verification feature points; Steps for establishing the motion model library: Establish the motion model library based on the verified motion information and the corresponding operation instructions. 3. The anti-cheating network survey method based on liveness detection as described in claim 2, wherein the model building step further includes a facial recognition step: constructing a facial recognition model library based on the obtained user facial recognition information. 4. The anti-cheating network survey method based on liveness detection as described in claim 3, characterized in that the feature comparison step specifically includes the following sub-steps: Similarity judgment steps: Determine whether the similarity between the action recognition information and the verification action information in the action recognition model library is greater than a preset value. If so, the verification is successful. 5. The anti-cheating network survey method based on liveness detection as described in claim 4, wherein the feature comparison step further includes a face comparison step: comparing the obtained face recognition information with the data in the face recognition model library; if the comparison results are consistent, then performing a similarity judgment step. 6. A cheating prevention network survey device based on liveness detection, characterized in that it includes the following modules: Questionnaire Setup Module: Questionnaire management users configure questionnaire content, survey question types, and matching user types through the questionnaire setup module, and then publish the questionnaire. The online questionnaire allows users to view questions, answer them, and submit information. The online questionnaire includes survey questions set by the questionnaire management user and randomly inserted user verification questions, or randomly selected true/false questions requiring users to nod for yes and shake their heads for no. User verification questions are randomly inserted questions involving specific facial expressions. Prompt Steps: Add a user verification step that combines liveness detection technology with prompts to complete facial movements, allowing users to perform specified facial movements according to the prompts; Model building module: used to build an action recognition model library; Information acquisition module: used to acquire the user's action recognition information, including the current feature vector of the human face; the system acquires video images of the user's specified facial actions as prompted by the camera, and extracts the user's facial feature vector from the video; Feature comparison module: This module compares the user's action recognition information with the verification feature vectors in the action recognition model library. Specifically, it compares the extracted facial feature vectors of the user in the video with the corresponding action feature vector set in the recognition model. If the comparison results match, the verification is passed. 7. The anti-cheating network survey device based on liveness detection as described in claim 6, characterized in that the model building module specifically includes the following sub-modules: Action acquisition module: used to acquire verification action information, the verification action information including the verification feature vector of the human face, the verification feature vector being the displacement change of the verification feature points; Action model library creation module: Used to create an action model library based on the verification action information and the corresponding operation instructions. 8. The anti-cheating network survey device based on liveness detection as described in claim 7, wherein the model building module further includes a facial recognition module: used to construct a facial recognition model library based on the acquired user's facial recognition information. 9. The anti-cheating network survey device based on liveness detection as described in claim 8, wherein the feature comparison module specifically includes the following sub-modules: Similarity judgment module: Used to determine whether the similarity between the action recognition information and the verification action information in the action recognition model library is greater than a preset value. If so, the verification is successful. 10. A cheat-prevention network survey system based on liveness detection, characterized in that it includes an actuator, the actuator being used to execute the cheat-prevention network survey method based on liveness detection as described in any one of claims 1-5.