A method for electronic examination paper review based on machine vision

Abstract

The present invention discloses a method for electronic examination paper review based on machine vision, characterized in that it comprises the following steps: S1, collecting the video stream displayed on the display screen of the examination equipment through a camera device, S2, slicing the video stream to generate video frames, inputting the video frames into a trained intelligent analysis model, the intelligent analysis model extracting the content in the video to output the candidate information and the examination process data, and forming an examination process data form, S3, comparing the examination process data form with the answer results in the examination system to verify the authenticity of the answer. The technology of the present invention can effectively improve the security of network platform examination supervision, especially can avoid the use of technical means to tamper with network data packets, or modify the background data to cheat.

Description

Technical Field

[0001] The invention relates to artificial intelligence and computer image processing foundations, and in particular to an electronic examination paper review method based on machine vision. Background Art

[0002] In order to ensure the fairness of the computer network platform examination, it is currently supervised by video.

[0003] Existing video monitoring uses computer vision technology to analyze the examinee's movements and then analyze possible cheating behaviors. For example, a video-based method and system for extracting abnormal frames in an examination room is disclosed in a Chinese invention patent (authorization publication number CN115830517B). The method obtains the first video frame of the examination room, generates examination room background data based on the first video frame, then captures multiple desktops from the examination room background data, and then obtains a differential image through a differential algorithm, and marks abnormal frames based on the edge data of the test paper and the upper limb data of the target object. The method presets reference objects such as desktops, active intervals, and prohibited intervals in the examination room background data, and determines multiple abnormal frames through the pixel relationship between the test paper and the target object and the reference object. This method can more accurately extract abnormal frames in the exam, greatly reducing the workload and providing convenience for subsequent identification of cheating behaviors.

[0004] However, the existing monitoring system and examination system are two independent systems that are not connected to each other. The current operating mode is: when reviewing the test papers of candidates in the examination system, if the reviewer finds that a certain test paper is suspected of cheating, the examinee's test process video will be retrieved from the monitoring system for browsing and checking to determine the authenticity of the test paper. Overall, the two systems are information islands and lack overall planning and coordination, which will inevitably cause a waste of time and resources and fail to meet the needs of coordinated management. In particular, when cheating is committed by tampering with network data packets or modifying background data through technical means, the existing technical means lack identification methods. Summary of the invention

[0005] The purpose of the present invention is to provide a safer electronic examination paper review method based on machine vision in view of the deficiencies in the prior art.

[0006] In order to achieve the above object, the technical solution adopted by the present invention is:

[0007] A method for electronic examination paper review based on machine vision comprises the following steps:

[0008] S1. Capture the video stream displayed on the display screen of the test equipment through a camera device.

[0009] S2, slicing the video stream to generate video frames, inputting the video frames into a trained intelligent analysis model, the intelligent analysis model extracting the content in the video and outputting the candidate information and the test process data to form a test process data form,

[0010] S3. Compare the test process data form with the answer results in the test system to verify the authenticity of the answers.

[0011] The preferred technical solutions of the present invention are as follows:

[0012] Preferably, the intelligent analysis model training method is as follows:

[0013] A1. Collect sample images of the answering interface.

[0014] A2. Input the collected image sample materials into the AI ​​open platform for algorithm training.

[0015] A3. Use the video frame to input the intelligent analysis model to optimize the model.

[0016] Preferably, the UI interface of the electronic test is first collected in step A1. If the accuracy of the training model in step A2 is insufficient, the UI interface in step A1 is adjusted, and the adjusted UI interface is used for answering questions and collecting data.

[0017] Preferably, in step S2, the intelligent analysis model calls the image data parsing algorithm package successfully trained by the AI ​​open platform from the algorithm management platform, reads the text information on the image, obtains the candidate information and the test process data respectively, and automatically generates a form.

[0018] Preferably, the candidate information includes the admission ticket number, and the test process data includes question numbers and options.

[0019] Preferably, slicing the video stream is to extract the video frames at 1 frame / s from the video stream, and store the video frames in a memory.

[0020] Preferably, in step S3, when the comparison results are inconsistent, the video frames corresponding to the inconsistent results in the memory are retrieved for manual verification.

[0021] Preferably, when the video frame cannot be manually checked, the corresponding video stream is retrieved for manual checking.

[0022] Preferably, step A1 includes adjusting the shooting angle of the camera device and the resolution of the display screen.

[0023] In summary, due to the adoption of the above-mentioned technical scheme, the beneficial effect of the present invention is: the technology of the present invention can effectively improve the security of network platform examination supervision, especially can avoid the use of technical means to tamper with network data packets, or modify background data to cheat. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 The flowchart of a specific embodiment of the present invention.

[0025] Figure 2 The image of the video stream is captured in a specific embodiment of the present invention.

[0026] Figure 3 It is a schematic diagram of the process of optimizing the model in a specific embodiment of the present invention.

[0027] Figure 4 The figure is a schematic diagram of the layout of the examination system in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0028] The present invention will be described in detail below in conjunction with the accompanying drawings.

[0029] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention. Example

[0030] like Figure 1 As shown, an embodiment of an electronic examination paper review method based on machine vision of the present invention includes the steps described below.

[0031] (1) Analysis of the UI interface of the examination system

[0032] The user provides the examination system account and password, and the algorithm trainer logs into the examination system to view the current electronic examination UI interface. After preliminary algorithm training, it is determined whether the UI interface needs to be modified to improve recognition accuracy. If the algorithm accuracy is insufficient, the examination system development staff will adjust the UI interface based on the training results and the suggestions of the algorithm trainer; if the algorithm accuracy meets the actual application requirements, the sample collection step will be entered.

[0033] (2) Sample collection

[0034] Algorithm trainers log into the examination system and collect a large amount of sample image materials including different question types.

[0035] (3) Algorithm training

[0036] Using the collected images as samples, the AI ​​open platform is used to train, remove duplicates, and optimize the algorithm. The trained algorithm is deployed to the hardware device at the center for invocation. The model is then continuously optimized based on the on-site data to form a closed-loop algorithm training process (such as Figure 3 as shown).

[0037] In the step, a specific AI open platform can be selected from the AI ​​open platform provided by https: / / ai.hikvision.com / , the sample image material collected in the above step (2) is imported into the AI ​​open platform, and then labels are manually established in the annotation, and then the images are annotated. After the annotation is completed, data quality diagnosis and training are performed. After the training is completed, the model is exported to obtain the algorithm.

[0038] After training and generating a new intelligent algorithm, the intelligent algorithm can be loaded into the algorithm warehouse for intelligent analysis. The above process can be iterated continuously, and the accuracy of the new algorithm can be continuously improved by increasing the number of training samples.

[0039] The algorithm management platform has complete algorithm life cycle management functions, including algorithm management, packaging, publishing, updating, cascading and other capabilities, which can effectively support users' flexible algorithm management and usage needs. The algorithm management platform is the core of realizing video intelligent analysis and value mining. The algorithm management platform manages the detection algorithm, modeling algorithm, comparison algorithm and other algorithm packages at the edge and center ends, provides a unified authorization system for algorithms, and provides a basis for the coordinated scheduling and application of algorithms.

[0040] The events of the front-end and back-end are strategically arranged, uniformly managed and executed to form a strategy cube. After the algorithm events are calculated by the strategy, the events with annotations are output, and the calculations and statistics are uniformly performed to output the effectiveness of the strategy application.

[0041] (4) Video cascading

[0042] In the electronic examination room, high-definition cameras have been deployed to shoot computer screens one by one, and manual focus is used to ensure that the answer interface appears as complete as possible on the monitoring screen (such as Figure 2 As shown in the figure), the video stream is fed into the existing video platform. The video resources of the established platform are accessed through platform networking, and the GB / T28181-2016 national standard protocol is used to achieve platform docking. The video resources of the subordinate platform are accessed through platform networking. The networking of the upper and lower platforms should meet the mandatory requirements of the "Technical Requirements for Information Transmission, Exchange and Control of Public Security Video Surveillance Networking Systems" (GB / T28181-2016). The principle of cascading docking of the upper and lower platforms is as follows: Figure 4 shown.

[0043] (5) Obtaining the test process data form

[0044] After the test begins, the video stream enters the intelligent analysis platform, which slices and captures it to generate images. At the same time, the trained algorithm is dispatched to collect the candidate information (name, gender, admission ticket number) and answer status (question number, answer) on the test interface to form a test process data form. Among them, the captured questions include true or false questions (options A / B) and single / multiple choice questions (options A / B / C / D). After the test, the answer file is formed according to the candidate information.

[0045] Based on the electronic test paper review algorithm training, optimization, and scheduling services, as well as video cascading, image capture and storage access services provided by the intelligent analysis platform, combined with actual application needs, the intelligent analysis platform can now provide business capabilities such as examination process data form acquisition, surveillance video management and captured image management.

[0046] Acquisition of test process forms: Through cascade access to existing video platforms and combined with the server's slice capture capabilities, the platform obtains the original video and captured images of the test process. The platform calls the image data parsing algorithm package successfully trained by the AI ​​open platform from the algorithm management platform, reads the text information on the image, and obtains the candidate information (admission ticket number, etc.) and test process data (question number and options), automatically generates a form, and pushes it to the upper-level platform through an open interface for further processing and archiving.

[0047] Snapshot image management: After the video stream of the existing video platform enters the intelligent analysis platform, the platform will capture 1 picture / s, store the captured pictures on the platform, and analyze and push them. The platform calls the successfully trained picture data analysis algorithm in the algorithm management platform to analyze the candidate information and test data in the picture to obtain the target test process form. At the same time, the platform stores the pictures. Then, through the open interface, the platform is pushed to the upper platform for business application.

[0048] (6) System connection

[0049] Provide an open interface for the intelligent analysis platform to connect with the examination system. After the examination, the candidate's answer file is transferred to the examination system and compared with the answer results in the system to verify the authenticity of the answer. At the same time, the original captured pictures and videos are retained in the examination analysis system. When the results are inconsistent, the original pictures and videos can be retrieved for manual review.

[0050] The test process data form and test process snapshots obtained from the intelligent analysis platform are pushed to the test business platform through an open interface and analyzed and compared with the test takers' answer results in the test system. If the comparison results are consistent, the test takers' test scores are valid; if the comparison results are inconsistent, the snapshots are retrieved for further analysis.

[0051] The original video is pushed to the evaluation center through the intelligent analysis platform. When the test results are in doubt and cannot be judged by the original pictures, the evaluation center staff can retrieve the original video and conduct manual verification to more accurately determine whether the examinee has cheated, thereby improving the credibility of the test.

[0052] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for electronic examination paper review based on machine vision, characterized in that: The following steps are involved: S1. Collect the video stream displayed on the display screen of the test equipment through the camera device; in the electronic examination room, arrange cameras to shoot the computer screen one by one, and through manual focus, ensure that the answer interface appears as complete as possible on the monitoring screen, and the video stream is transmitted to the existing video platform; Access the video resources of the established platform through platform networking, and use the GB / T28181-2016 national standard protocol to achieve platform connection; access the video resources of the lower-level platform through platform networking; After the question-answering begins, the video stream enters the video platform; S2, slicing the video stream to generate video frames, inputting the video frames into a trained intelligent analysis model, the intelligent analysis model extracting the content in the video and outputting the candidate information and the test process data to form a test process data form, S3. Compare the test process data form with the answer results in the test system to verify the authenticity of the answers; the intelligent analysis model training method is as follows: A1. Collect sample images of the answering interface. A2. Input the collected image sample materials into the AI ​​open platform for algorithm training. A3, using the video frame to input the intelligent analysis model to optimize the model; In step S2, the intelligent analysis model calls the image data analysis algorithm package successfully trained on the AI ​​open platform from the algorithm management platform, reads the text information on the image, obtains the candidate information and the test process data respectively, and automatically generates a form; The candidate information includes the admission ticket number, and the test process data includes the question number and options; It also includes: analyzing and comparing the examinee's examinee information and examination process data with the examinee's answer results in the examination system. If the comparison results are consistent, the examinee's examination results are valid; if the comparison results are inconsistent, retrieving the captured images for further analysis; In step A1, the UI interface of the electronic test is first collected. If the accuracy of the training model in step A2 is insufficient, the UI interface in step A1 is adjusted, and the adjusted UI interface is used for answering questions and collecting data; Slicing the video stream is to extract the video frames at a rate of 1 frame / s from the video stream, and store the video frames in a memory.

2. The method for electronic examination paper review based on machine vision according to claim 1, characterized in that: In step A3, when the comparison results are inconsistent, the video frames corresponding to the inconsistent results in the memory are retrieved for manual verification.

3. The method for electronic examination paper review based on machine vision according to claim 2, characterized in that: When the video frame cannot be manually checked, the corresponding video stream is retrieved for manual checking.

4. The method for electronic examination paper review based on machine vision according to claim 1, characterized in that: Step A1 includes adjusting the shooting angle of the camera device and the resolution of the display screen.

Images