A video condensation method based on national education examinations

Abstract

The present invention relates to the technical field of video condensation, and discloses a video condensation method based on national education examinations, including the following steps: S1. Detect the number of people, the area of scene change, and the number of scene change regions in the video to be condensed, and record the results as: M, Image binary , N; S2. Set the three parameters of M, Image binary , N according to the contribution to 60%, 20%, and 20%, integrate and quantify the score of the condensed video into the space of 0-100, and obtain the unified quantization value f s of the three parameters. The calculation formula is: f s = α * 0.6 + β * 0.2 + γ * 0.2. The video condensation method based on national education examinations can compress the video file in terms of time and space, and improve the playback efficiency and quality of the user's file.

Description

Technical Field

[0001] The present invention relates to the technical field of video condensation, and specifically provides a video condensation method based on national education examinations. Background Art

[0002] According to the "Technical Specification Standard for National Education Examination Online Patrol System (2017 Edition)", inspection centers at all levels should store audio and video during the examination for at least half a year. Currently, the examination institute requires the video recording in the security room to be played back every 6 hours, and for the videos in the examination rooms, spot checks will be conducted. However, the playback of the recordings is all carried out by manual multiple playback, resulting in low efficiency.

[0003] Currently, the condensation solutions on the market analyze the moving targets in the video through algorithms, extract the moving targets, then analyze the movement trajectories of each target, splice different targets into a common background scene, and combine them in a certain way to generate a new condensed video. This solution is for tracking targets and is not applicable to the situation where there are a large number of targets in the examination room for a long time during the examination. Summary of the Invention

[0004] (I) Technical Problems to be Solved

[0005] Aiming at the deficiencies of the prior art, the present invention provides a video condensation method based on national education examinations, which has the advantages of judging the value of images by combining the use of human detection, scene change area, and the number of scene change regions, and solves the problem of low efficiency in the prior art where the playback of examination video recordings is all carried out by manual multiple playback.

[0006] (II) Technical Solutions

[0007] To achieve the purpose of improving the quality and efficiency of spot checks on examination invigilation videos, the present invention provides the following technical solutions: A video condensation method based on national education examinations, including the following steps:

[0008] S1. Detect the number of people, scene change area, and the number of scene change regions in the video to be condensed, and record the results as: M, Image binary , N;

[0009] S2. Set the three parameters M, Image binary , N according to their contributions as 60%, 20%, and 20%, and integrate and quantify the score of the condensed video into the space of 0 - 100. Specifically:

[0010] b. Quantification of human detection results: When M ≥ 10, no condensation is performed. When M < 10, the quantization value α of human detection is obtained according to the following formula:

[0011]

[0012] b. Quantification of the scene change area detection result: According to the detection result Image binary , count the points with the image value of 255, and obtain the quantified value β of the scene change area according to the following formula:

[0013] β = 100 * Area binary / (w * h)

[0014] where w and h represent the width and height of the image respectively;

[0015] c. Quantification of the scene change region quantity detection result: If N ≥ 10, no concentration is performed. If N < 10, obtain the quantified value γ of the scene change region quantity according to the following formula:

[0016]

[0017] d. Integrate and calculate α, β, and γ according to the following formula to obtain the unified quantified value f of the three parameters s :

[0018] f s = α * 0.6 + β * 0.2 + γ * 0.2

[0019] S3. Set a default value of f s . If the actual value of f s is lower than the default value, concentration is performed.

[0020] Furthermore, in step S1, when detecting the number of people in the video to be concentrated, the yolov4 detection model under the darknet framework is adopted.

[0021] Furthermore, in step S1, when detecting the scene change area in the video to be concentrated, use the GMM background modeling method to establish a Gaussian mixture model for each pixel in a video sequence frame to obtain the binary foreground image Image binary .

[0022] Furthermore, in step S1, when detecting the scene change area in the video to be concentrated, it is to count the connected regions in Image binary , and the number of connected regions is denoted as N.

[0023] Furthermore, in step S2, the larger the values of α, β, and γ, the higher the concentration degree.

[0024] Furthermore, before quantifying the scene change region quantity detection result in step c, it is necessary to determine whether N meets the standard of scene region change. When determining, according to the calculation of Imagebinary Count the number of pixels N in each connected region where the pixel value is 255 respectively k If N k is greater than the set value N max then the connected region meets the standard of scene region change

[0025] (III) Beneficial effects

[0026] Compared with the prior art, the present invention provides a video condensation method based on national education examinations, having the following beneficial effects

[0027] 1. The video condensation method based on national education examinations, based on the video condensation method for national education examinations, uses person detection and scene change detection to perform structured analysis on the examination room video, regenerates video tags for valuable data, completes the compression of the video file in terms of space and time, is conducive to the preservation of the video, and improves the quality and efficiency of subsequent spot checks

[0028] 2. The video condensation method based on national education examinations, for the convenience of users, uses three parameters of the number of people, the area of scene change, and the scene change region, and sets different function scores, unifies these scores and quantifies them between 0 - 100. When f s is lower than the set default value, the video value is low and can be condensed. The quantified processing method reaches an intuitive and effective evaluation standard Specific implementation manners

[0029] The technical solutions in the embodiments of the present invention will be clearly and completely described below. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention

[0030] A video condensation method based on national education examinations includes the following steps

[0031] S1. Detect the number of people in the video to be condensed. The detection of the number of people uses the yolov4 detection model under the darknet framework. First, collect the images containing people in the examination room, and then perform the annotation work of the human heads through the annotation tool. During the annotation process, only the human heads need to be framed, and then the annotated data is trained on the GPU server to obtain a trained model. Finally, use this model to perform the work of human head detection, and the number of detected human heads is M

[0032] S2. Detect the area of scene change in the condensed video: Use the GMM background modeling method to establish a Gaussian mixture model for each pixel in a video sequence frame. In this model, the Gaussian model with a large weight represents the background, and the one with a small weight represents the foreground. If a new pixel can match the Gaussian model corresponding to the background, then the new pixel is regarded as the background. If it matches the Gaussian model with a smaller weight or there is no matching Gaussian model, then this pixel is regarded as the foreground, and the binary foreground image Image is obtained. binary ;

[0033] S3. Detect the number of scene change regions in the condensed video: Count the connected regions in Image binary and record the number of connected regions, which is N.

[0034] S4. Set the three parameters M, Image binary , and N as 60%, 20%, and 20% according to their contributions, and integrate and quantize the score of the condensed video into the space of 0 - 100. Specifically:

[0035] a. Quantization of the person detection result: When M ≥ 10, no condensation is performed. When M < 10, the quantization value α of person detection is obtained according to the following formula:

[0036]

[0037] b. Quantization of the scene change area detection result: According to the detection result Image binary , count the points with the image value of 255, and obtain the quantization value β of the scene change area according to the following formula:

[0038] β = 100 * Area binary / (w * h)

[0039] where w and h represent the width and height of the image respectively.

[0040] c. Quantization of the scene change region number detection result: When N ≥ 10, no condensation is performed. When N < 10, the quantization value γ of the scene change region number is obtained according to the following formula:

[0041]

[0042] d. Integrate and calculate α, β, and γ according to the following formula to obtain the unified quantization value f of the three parameters s :

[0043] f s = α * 0.6 + β * 0.2 + γ * 0.2S3. Set a default value of f s When f sIf the actual value is lower than the default value, concentration is performed.

[0044] It should be noted that before quantifying the detection result of the number of scene change regions in step c, it is necessary to determine whether N meets the standard of scene region change. When making the determination, calculate Image binary and respectively count the number of pixels N of each connected region with a pixel value of 255 k , if N k is greater than the set value N max , then the connected region meets the standard of scene region change.

[0045] The above video concentration method based on national education examinations uses the combined quantization results of three parameters: person detection, scene change area, and number of scene change regions to determine whether an image is valuable, thereby determining the value of the video. If it is lower than the set f s default value, it has no value. For national education examinations, when the examination room is in the non - use stage and there is no test paper access in the security room, most of the images are static and unchanged. The video concentration method in this application can greatly compress the time and space of video files for such video files, improving the work quality and efficiency of users.

[0046] It should be noted that the term "comprising", "including" or any other variant thereof is intended to cover non - exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements but also other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising one..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the said element.

[0047] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A video condensation method based on national education examinations, characterized in that , including the following steps: S1. Detect the number of people, the area of scene change, and the number of scene change regions in the condensed video, and record the results as M, Image binary , N; S2. Set the three parameters M, Image binary , and N according to their contributions as 60%, 20%, and 20%, and integrate and quantify the score of the condensed video into the range of 0 - 100, specifically as follows: a. Quantification of the person detection result: When M ≥ 10, no concentration is performed. When M < 10, the quantification value α of the person detection is obtained according to the following formula: b. Quantification of the scene change area detection result: Based on the detection result Image binary , for this foreground image Image binary count the points with an image value of 255, and obtain the quantified value β of the scene change area according to the following formula: β = 100 * Image binary / (w * h) where w and h represent the width and height of the image respectively; c. Quantification of the detection result of the number of scene change regions: When N ≥ 10, no concentration is performed. When N < 10, the quantification value γ of the number of scene change regions is obtained according to the following formula: d. Integrate and calculate α, β, and γ according to the following formula to obtain the unified quantization value f of the three parameters s :[[-END]] f s = α * 0.6 + β * 0.2 + γ * 0.2 S3. Set a default value for f s and perform concentration when the actual value of f s is lower than the default value.

2. The video condensation method based on national education examinations according to claim 1, wherein , in step S1, when detecting the number of people in the video to be concentrated, the yolov4 detection model under the darknet framework is used.

3. A video condensation method based on national education examinations according to claim 1, characterized in that , In step S1, when detecting the scene change area in the video to be concentrated, the Gaussian Mixture Model (GMM) background modeling method is used to establish a mixture Gaussian model for each pixel in a video sequence frame, and the foreground image Image after binarization is obtained. binary .

4. A video condensation method based on national educational examinations according to claim 1, characterized in that , in step S1, when detecting the scene change area in the video to be concentrated, the connected regions in Image binary are counted, and the number of connected regions is denoted as N.

5. A video condensation method based on national educational examinations according to claim 1, characterized in that , in step S2, the larger the values of α, β, and γ, the higher the concentration degree.

6. The video condensation method based on national educational examinations according to claim 1, characterized in that , Before quantifying the detection result of the number of scene change regions in step c, it is necessary to determine whether N meets the standard of scene region change. When making the determination, calculate Image binary to count the number of pixels N of each connected region with a pixel value of 255 respectively k , if N k is greater than the set value N max , then the connected region meets the standard of scene region change.

Images