CN117478838B - A distributed video processing supervision system and method based on information security - Google Patents

Abstract

The invention discloses a distributed video processing supervision system and method based on information security, and belongs to the technical field of information security. The system comprises a data acquisition module, a data analysis module, a risk identification module and an operation management module; the data acquisition module is used for acquiring the operation parameters of the live broadcast equipment and the shot video; the data analysis module is used for analyzing the video information frame by frame, dividing key areas by combining the operation parameters of the equipment, identifying target objects, calculating important indexes of the target objects, and determining video subjects according to the important indexes; the risk identification module is used for judging the association degree of the target object and the video theme, substituting the association degree into a formula to calculate the risk index of each target object, and finding out the risk object to be monitored according to the risk index; the operation management module adopts a distributed technology to control a plurality of devices to monitor the areas where different risk objects are located respectively, and timely performs image processing when illegal contents appear, so that information safety is protected.

Description

A distributed video processing supervision system and method based on information security

Technical Field

The present invention relates to the field of information security technology, and in particular to a distributed video processing supervision system and method based on information security.

Background technique

With the rapid development of mobile Internet, live video has become a popular form of media. People can watch live content anytime and anywhere through mobile phones, computers and other devices. Live video covers various fields, including entertainment, sports, education, news, etc., and has become an important channel for people to obtain information and entertainment. However, live video also faces some problems and challenges. First, the real-time requirements of live video are high, and the video signal on the scene needs to be quickly transmitted to the audience end to ensure that the audience can watch the live content in real time. Secondly, live video usually needs some processing, such as image enhancement, noise removal, target detection, etc., to improve video quality and viewing experience. In addition, there may be some sensitive information in the live video, such as personal privacy or commercial secrets, which needs to be protected and processed to ensure information security.

At present, information security issues that may arise during live broadcasts are usually solved by manual observation or intelligent algorithm recognition. Manual observation means that during the live broadcast process, there is a dedicated person responsible for observing the live broadcast environment, reminding or controlling the adjustment of the live camera shooting picture. This method has certain defects. Since human reaction takes a certain amount of time, and the live broadcast environment is ever-changing, when sensitive information is leaked, people often cannot react in time to deal with it. The use of delayed live broadcast pictures to leave enough time for manual review will affect the real-time nature of the live broadcast and bring a bad experience to the audience. Compared with manual observation, intelligent algorithm recognition is more efficient. Intelligent algorithm refers to the use of computer vision technology to automatically detect and block sensitive information through real-time analysis and recognition of videos. Compared with manual observation, this method can greatly improve detection efficiency and processing speed, but it also has inevitable defects. For example: 1. Lack of flexibility in identifying sensitive information through manual observation methods: It is impossible to adjust the type of sensitive information detection for different live broadcast environments, and it can only be mechanically screened based on the existing comparison library; 2. Lack of sufficiently delicate image processing capabilities: Due to the high real-time requirements of live broadcast videos, there is usually a certain time requirement from recording to processing and then sending the video to the network. In such a short period of time, a single device cannot provide computing power support to meet pixel-level image processing. Even when the performance of the device is poor, even static image-level image processing cannot be completed. It can only extract frames or directly block processing in fixed areas of the video, resulting in a significant decrease in the live broadcast effect experience. Therefore, at this stage, a more flexible and efficient technical solution for information security detection and processing is needed to solve the above problems.

Summary of the invention

The purpose of the present invention is to provide a distributed video processing supervision system and method based on information security to solve the problems raised in the above background technology.

In order to solve the above technical problems, the present invention provides the following technical solutions: a distributed video processing supervision method based on information security, the method comprising the following steps:

S1. Real-time collection of operating parameters of live broadcast equipment and real-time video, and acquisition of the latest theme library;

S2, dividing the video screen into regions by analyzing the operating parameters, identifying the target object and calculating the important index;

S3, select the subject object according to the important index and determine the relevance of each target object;

S4. Use distributed technology to monitor and process images of target objects with different associations.

In S1, the operating parameters refer to the focal length and shooting distance of the live broadcast device camera. The shooting distance refers to the distance between the camera and the shooting target, and the distance sensor installed inside the live broadcast device collects data. The theme library is used as a comparison library for live broadcast types to identify the type of each live broadcast. The theme library contains theme collections of various live broadcast types, and the theme collection includes the names of different items under the same theme.

In order to improve the coverage of live broadcast type recognition, the theme library should contain a sufficient number of theme sets when it is established to meet the recognition requirements of different live broadcast types. The name of the type of items in each theme set should also cover as many types of items as possible that may appear during the live broadcast of the corresponding live broadcast type.

In S2, the specific steps are as follows:

S201, obtaining the real-time video captured by the live broadcast device, using the existing OpenCV to decompose the video into single-frame images, screening all the single-frame images, and using the edge detection method to remove blurred single-frame images.

S202, analyze the remaining single-frame images one by one, obtain the focal length and shooting distance of the live broadcast device camera at the corresponding time of each single-frame image, substitute them into the formula, and calculate the weight ratio R of each single-frame image respectively. Taking the center point of the single-frame image as the center, take the product of the length of the single-frame image and the weight ratio R as the length, and the product of the width of the single-frame image and the weight ratio R as the width, and divide a rectangular area as the key area; the weight ratio calculation formula is as follows:

Where R is the weight ratio, α is the focal length influence coefficient, j is the focal length, J _max is the maximum focal length of the device, β is the shooting distance influence coefficient, a is the distance constant, and P is the shooting distance.

S203, using the YOLOv3 target detection algorithm to detect all target objects in each single-frame image and assign identifiers, extracting feature information of the target objects and analyzing their types and positions, and calculating the number of pixels occupied by the target objects, marking target objects located in key areas as key areas, and marking other target objects as ordinary.

S204, obtaining the position coordinates ( _xi , _yi ) of the key target object and the position coordinates ( _xz , _yz ) of the center point of the corresponding single-frame image, and substituting the number of pixels of the key target object into the formula to calculate the importance index, each key target object corresponds to an importance index; the calculation formula is as follows:

Where ZY is the importance index, _Si is the number of pixels of the key target object, _Sz is the number of pixels in the key area of a single frame image, u is the distance influence coefficient, and c is the position constant.

The location coordinates of the target object are the center coordinates of the area where it is located, and the number of pixels is the number of pixels in the area where it is located. The more pixels there are, the larger the coverage area of the area where it is located.

S205. Perform similarity recognition between target objects in different single-frame images, associate target objects whose similarity is higher than a similarity threshold and do not belong to the same single-frame image, sum the importance indexes of the mutually associated key target objects to obtain a total importance index, and use the mutually associated target objects as the embodiment of the same target object in different single-frame images, and unify the identifiers of all mutually associated target objects.

In S3, the specific steps are as follows:

S301. The key target objects whose total importance index is greater than the importance index threshold are taken as subject objects, and the names of the categories to which all subject objects belong are put into a subject category set, which includes {Q ₁ , Q ₂ , Q ₃ , ..., Q _n }, where n represents the number of category names and Q _n represents the nth category name.

S302. Compare all category names in the subject category set with all category names in each subject set in the subject library, determine whether the category names in the subject set are the same as the category names in the subject category set, and if they are the same, mark the category names in the corresponding subject set. After all subject sets in the subject library have completed the comparison and judgment, rank the subject sets in ascending order according to the number of marked category names, and select the subject set ranked first as the current subject set.

S303. Obtain the category names of all target objects, and compare them with all category names in the current subject set respectively to determine whether the same category names exist in the current subject set. If so, set the association degree of the corresponding target object to strong association. If not, continue to compare with all category names in other subject sets in the subject library except the current subject set to determine whether the same category names exist. If so, set the association degree of the corresponding target object to weak association. If not, set the association degree of the corresponding target object to no association.

In S4, the specific steps are as follows:

S401. Set different influence coefficients for different associations. The influence coefficients are set in the order of no association, weak association, and strong association from large to small. Obtain the association of all target objects, and substitute the influence coefficient and the number of pixels corresponding to the association of the target object into the formula to calculate the risk index. The formula is as follows:

FX＝K×E×S

Where FX is the risk index of the target object, K is the influence coefficient, E is a constant, and S is the number of pixels of the target object.

S402. Obtain the number of callable devices I in the distributed devices, rank the target objects in positive order according to the risk index, select the top I target objects as risk objects, and assign devices to these risk objects in turn. The assigned devices monitor the areas where the risk objects are located.

S403. When the risk object is monitored to be an illegal item or an illegal behavior occurs, pixel-level image processing is performed on the area where the risk object is located without affecting the overall clarity of the single-frame image. The processing methods include mosaic pixelization, pixel repair method, and pixel replacement method.

Mosaic pixelation: Mosaic the illegal area and pixelate it to achieve the effect of hiding. Pixel repair method: Repair the illegal part and repair it into pixels that meet the requirements to achieve the effect of correction. Pixel replacement method: Replace the pixels of the illegal part with pixels that meet the requirements to achieve the effect of correction.

A distributed video processing and supervision system based on information security includes a data acquisition module, a data analysis module, a risk identification module and an operation management module.

The data acquisition module is used to collect the operating parameters of the live broadcast equipment and the real-time video taken; the data analysis module analyzes the video information frame by frame, divides the key areas and identifies the target objects based on the operating parameters of the equipment, calculates the importance index of the target objects, and determines the video theme based on the importance index; the risk identification module is used to determine the correlation between the target object and the video theme, substitutes the correlation into the formula to calculate the risk index of each target object, and finds out the risk objects that need to be monitored based on the risk index; the operation management module uses distributed technology to control different devices to monitor the area where each risk object is located, and performs image processing when the risk object is an illegal item or an illegal behavior occurs to protect information security.

The data acquisition module includes an image information acquisition unit, a device information acquisition unit and a subject information acquisition unit.

The image information acquisition unit is used to collect real-time video captured by the live broadcast device. The device information acquisition unit is used to collect the operating parameters of the live broadcast device when it is working. The operating parameters include focal length and shooting distance. The shooting distance refers to the distance between the live broadcast device and the shooting target. The theme information acquisition unit is used to collect the theme library in the system. The theme library contains theme collections of various live broadcast types, and the theme collection includes the type names of different items under the same theme.

The data analysis module includes a region division unit, an object recognition unit and a subject analysis unit.

The area division unit is used to divide the key areas in the video screen. First, the real-time video shot by the live broadcast device is decomposed into single-frame images; second, the operating parameters of the live broadcast device at the corresponding time of each single-frame image are obtained, and the weight ratio of each single-frame image is calculated in the formula; finally, with the center point of the single-frame image as the center, the product of the length of the single-frame image and the weight ratio R is used as the length, and the product of the width of the single-frame image and the weight ratio R is used as the width, and a rectangular area is divided as the key area. Each single-frame image has a key area.

During the live broadcast process, the live broadcast object is usually located in the center area of the picture, and the size of the center area of the picture is directly and inversely proportional to the focal length and the shooting distance.

Proportional relationship: The larger the focal length, the closer the screen is to the subject, and the larger the center area of the screen should be to reasonably include the live object.

Inverse relationship: the larger the shooting distance is, the farther the screen is from the subject, and the smaller the center area of the screen should be to reasonably include the live object.

The object recognition unit is used to identify key target objects. The YOLOv3 target detection algorithm is used to detect the target object in each single frame image, extract the feature information of the target object, analyze its type and position, and calculate the number of pixels occupied by the target object. The target object located in the key area is regarded as the key target object.

The topic analysis unit is used to analyze the video topic.

First, the position coordinates of the key target object and the position coordinates of the center point of the corresponding single-frame image are obtained, and the number of pixels of the key target object is substituted into the formula to calculate the importance index.

Secondly, the same target object in different single-frame images is associated, and the importance indexes of the associated key target objects are summed up to obtain the total importance index, and the names of the key target objects whose total importance index is greater than the importance index threshold are put into the subject category set.

Finally, all the category names in the subject category set are compared with all the category names in each subject set in the subject library to determine whether the category names in the subject set are the same as the category names in the subject category set. If they are the same, mark the category names in the corresponding subject set. After all the subject sets in the subject library have completed the comparison and judgment, rank the subject sets in ascending order according to the number of marked category names, and select the subject set ranked first as the current subject set.

The risk identification module includes a relevance judgment unit and a risk judgment unit.

The relevance judgment unit is used to judge the relevance of each target object with the video theme. The category name of the target object is compared with all category names in the current theme set to determine whether the same category name exists in the current theme set. If so, the relevance of the corresponding target object is set to strong relevance. If not, it continues to be compared with all category names in other theme sets in the theme library except the current theme set to determine whether the same category name exists. If so, the relevance of the corresponding target object is set to weak relevance. If not, the relevance of the corresponding target object is set to no relevance.

The risk judgment unit is used to calculate the risk index of each target object. Different influence coefficients are set for different associations. The influence coefficients are set in the order of no association, weak association and strong association from large to small. The influence coefficient and the number of pixels corresponding to the association degree of the target object are substituted into the formula to calculate the risk index.

A high correlation coefficient means that the target object is not closely related to the subject of the video, and the probability of a violation is higher. A larger number of pixels means that the target object's image accounts for a higher proportion of the entire image, and the impact of a violation is more severe. Comprehensively considering the correlation coefficient and the number of pixels can better assess the possible risks of the target object.

The operation management module is used to supervise risk objects. First, the number of callable devices in the distributed devices is obtained, and the target objects are ranked in positive order according to the risk index; second, the top I target objects are selected as risk objects, and devices are assigned to these risk objects in turn. The assigned devices monitor the areas where the risk objects are located; finally, when the risk objects are monitored as illegal items or illegal behaviors, pixel-level image processing is performed on the areas where the risk objects are located without affecting the overall clarity of the single-frame image.

Each device only monitors one risk object, and multiple devices can monitor and process multiple risk objects in a single frame image.

When the system is running, real-time video is continuously decomposed into single-frame images. After key area division and video subject identification, risk objects are quickly identified and distributed technology is used to use multiple devices to supervise the areas where different risk objects are located.

When some risk objects violate regulations, the computing power resources of multiple devices are used to perform fast image processing at the pixel level. The image processing is only for the area where the risk objects are located. After the processing is completed, the image of the processed area is quickly combined into the original single-frame image.

Compared with the prior art, the beneficial effects achieved by the present invention are:

1. Rapidly identify video themes: The present invention can quickly find the subject object by dividing the key areas and calculating the key index. Combined with the comparative analysis of the subject library, it can quickly locate the video subject and improve the flexibility and accuracy of information security detection.

2. Accurately locate risk objects: This application comprehensively considers the correlation influence coefficient and the number of pixels in calculating the risk index of the target object. The correlation influence coefficient represents the probability of violation of the target object, and the number of pixels represents the consequence of the violation of the target object. Target objects with high correlation influence coefficient and large number of pixels are given priority for monitoring, which can not only improve the efficiency of information security monitoring, but also reduce the risk of sensitive information leakage.

3. Pixel-level image processing: This application uses distributed technology to call multiple devices to supervise each risk object separately. When a violation occurs, the corresponding device only needs to perform pixel-level image processing on part of the area on each static image transmitted. The task is simple, consumes low computing resources and does not take too much time. The processed pixel areas are combined into a new single-frame image and sent to the network, which improves the level of detail of image processing without affecting the real-time performance of the live video.

In summary, compared with traditional technologies, the present invention has the advantages of rapid identification of video subjects, accurate positioning of risk objects and pixel-level image processing, and can improve the flexibility and efficiency of information security monitoring.

BRIEF DESCRIPTION OF THE DRAWINGS

The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

FIG1 is a flow chart of a distributed video processing supervision method based on information security according to the present invention;

FIG. 2 is a schematic diagram of the structure of a distributed video processing and monitoring system based on information security according to the present invention.

Detailed ways

The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

Referring to FIG. 1 , the present invention provides a distributed video processing supervision method based on information security, the method comprising the following steps:

S1. Real-time collection of operating parameters of live broadcast equipment and real-time video, and acquisition of the latest theme library;

S2, dividing the video screen into regions by analyzing the operating parameters, identifying the target object and calculating the importance index;

S3, select the subject object according to the important index and determine the relevance of each target object;

S4. Use distributed technology to monitor and process images of target objects with different associations.

In S1, the operating parameters refer to the focal length and shooting distance of the camera of the live broadcast device. The shooting distance refers to the distance between the camera and the shooting target, and the distance sensor installed inside the live broadcast device collects data. The theme library is used as a comparison library for live broadcast types to identify the type of each live broadcast. The theme library contains theme collections of various live broadcast types, and the theme collection includes the names of different items under the same theme.

In order to improve the coverage of live broadcast type recognition, the theme library should contain a sufficient number of theme sets when it is established to meet the recognition requirements of different live broadcast types. The name of the type of items in each theme set should also cover as many types of items as possible that may appear during the live broadcast of the corresponding live broadcast type.

In S2, the specific steps are as follows:

S201, obtaining the real-time video captured by the live broadcast device, using the existing OpenCV to decompose the video into single-frame images, screening all the single-frame images, and using the edge detection method to remove blurred single-frame images.

S202, analyze the remaining single-frame images one by one, obtain the focal length and shooting distance of the live broadcast device camera at the corresponding time of each single-frame image, substitute them into the formula, and calculate the weight ratio R of each single-frame image respectively. Taking the center point of the single-frame image as the center, take the product of the length of the single-frame image and the weight ratio R as the length, and the product of the width of the single-frame image and the weight ratio R as the width, and divide a rectangular area as the key area; the weight ratio calculation formula is as follows:

Where R is the weight ratio, α is the focal length influence coefficient, j is the focal length, J _max is the maximum focal length of the device, β is the shooting distance influence coefficient, a is the distance constant, and P is the shooting distance.

S203, using the YOLOv3 target detection algorithm to detect all target objects in each single-frame image and assign identifiers, extracting feature information of the target objects and analyzing their types and positions, and calculating the number of pixels occupied by the target objects, marking target objects located in key areas as key areas, and marking other target objects as ordinary.

S204, obtaining the position coordinates ( _xi , _yi ) of the key target object and the position coordinates ( _xz , _yz ) of the center point of the corresponding single-frame image, and substituting the number of pixels of the key target object into the formula to calculate the importance index, each key target object corresponds to an importance index; the calculation formula is as follows:

Where ZY is the importance index, _Si is the number of pixels of the key target object, _Sz is the number of pixels in the key area of a single frame image, u is the distance influence coefficient, and c is the position constant.

The location coordinates of the target object are the center coordinates of the area where it is located, and the number of pixels is the number of pixels in the area where it is located. The more pixels there are, the larger the coverage area of the area where it is located.

S205. Perform similarity recognition between target objects in different single-frame images, associate target objects whose similarity is higher than a similarity threshold and do not belong to the same single-frame image, sum the importance indexes of the mutually associated key target objects to obtain a total importance index, and use the mutually associated target objects as the embodiment of the same target object in different single-frame images, and unify the identifiers of all mutually associated target objects.

In S3, the specific steps are as follows:

S301. The key target objects whose total importance index is greater than the importance index threshold are taken as subject objects, and the names of the categories to which all subject objects belong are put into a subject category set, which includes {Q ₁ , Q ₂ , Q ₃ , ..., Q _n }, where n represents the number of category names and Q _n represents the nth category name.

S302. Compare all the category names in the subject category set with all the category names in each subject set in the subject library, and determine whether the category names in the subject set are the same as the category names in the subject category set. If they are the same, mark the category names in the corresponding subject set. After all the subject sets in the subject library have completed the comparison and judgment, rank the subject sets in ascending order according to the number of marked category names, and select the subject set ranked first as the current subject set.

S303. Obtain the category names of all target objects, and compare them with all category names in the current subject set respectively to determine whether the same category names exist in the current subject set. If so, set the association degree of the corresponding target object to strong association. If not, continue to compare with all category names in other subject sets in the subject library except the current subject set to determine whether the same category names exist. If so, set the association degree of the corresponding target object to weak association. If not, set the association degree of the corresponding target object to no association.

In S4, the specific steps are as follows:

S401. Set different influence coefficients for different associations. The influence coefficients are set in the order of no association, weak association, and strong association from large to small. Obtain the association of all target objects, and substitute the influence coefficient and the number of pixels corresponding to the association of the target object into the formula to calculate the risk index. The formula is as follows:

FX＝K×E×S

Where FX is the risk index of the target object, K is the influence coefficient, E is a constant, and S is the number of pixels of the target object.

S402. Obtain the number of callable devices I in the distributed devices, rank the target objects in positive order according to the risk index, select the top I target objects as risk objects, and assign devices to these risk objects in turn. The assigned devices monitor the areas where the risk objects are located.

S403. When the risk object is monitored to be an illegal item or an illegal behavior occurs, pixel-level image processing is performed on the area where the risk object is located without affecting the overall clarity of the single-frame image. The processing methods include mosaic pixelization, pixel repair method, and pixel replacement method.

Mosaic pixelation: Mosaic the illegal area and pixelate it to achieve the effect of hiding. Pixel repair method: Repair the illegal part and repair it into pixels that meet the requirements to achieve the effect of correction. Pixel replacement method: Replace the pixels of the illegal part with pixels that meet the requirements to achieve the effect of correction.

Please refer to FIG. 2 . The present invention provides a distributed video processing and supervision system based on information security. The system includes a data acquisition module, a data analysis module, a risk identification module and an operation management module.

The data acquisition module is used to collect the operating parameters of the live broadcast equipment and the real-time video taken; the data analysis module analyzes the video information frame by frame, divides the key areas and identifies the target objects based on the operating parameters of the equipment, calculates the importance index of the target objects, and determines the video theme based on the importance index; the risk identification module is used to determine the correlation between the target object and the video theme, substitutes the correlation into the formula to calculate the risk index of each target object, and finds out the risk objects that need to be monitored based on the risk index; the operation management module uses distributed technology to control different devices to monitor the area where each risk object is located, and performs image processing when the risk object is an illegal item or an illegal behavior occurs to protect information security.

The data acquisition module includes an image information acquisition unit, a device information acquisition unit and a subject information acquisition unit.

The image information acquisition unit is used to collect real-time video captured by the live broadcast device. The device information acquisition unit is used to collect the operating parameters of the live broadcast device when it is working. The operating parameters include focal length and shooting distance. The shooting distance refers to the distance between the live broadcast device and the shooting target. The theme information acquisition unit is used to collect the theme library in the system. The theme library contains theme collections of various live broadcast types, and the theme collection includes the type names of different items under the same theme.

The data analysis module includes a region division unit, an object recognition unit and a subject analysis unit.

The area division unit is used to divide the key areas in the video screen. First, the real-time video shot by the live broadcast device is decomposed into single-frame images; second, the operating parameters of the live broadcast device at the corresponding time of each single-frame image are obtained, and the weight ratio of each single-frame image is calculated in the formula; finally, with the center point of the single-frame image as the center, the product of the length of the single-frame image and the weight ratio R is used as the length, and the product of the width of the single-frame image and the weight ratio R is used as the width, and a rectangular area is divided as the key area. Each single-frame image has a key area.

During the live broadcast process, the live broadcast object is usually located in the center area of the picture, and the size of the center area of the picture is directly and inversely proportional to the focal length and the shooting distance.

Proportional relationship: The larger the focal length, the closer the screen is to the subject, and the larger the center area of the screen should be to reasonably include the live object.

Inverse relationship: the larger the shooting distance is, the farther the screen is from the subject, and the smaller the center area of the screen should be to reasonably include the live object.

The object recognition unit is used to identify key target objects. The YOLOv3 target detection algorithm is used to detect the target object in each single frame image, extract the feature information of the target object, analyze its type and position, and calculate the number of pixels occupied by the target object. The target object located in the key area is regarded as the key target object.

The topic analysis unit is used to analyze the topic of the video.

First, the position coordinates of the key target object and the position coordinates of the center point of the corresponding single-frame image are obtained, and the number of pixels of the key target object is substituted into the formula to calculate the importance index.

Secondly, the same target object in different single-frame images is associated, and the importance indexes of the associated key target objects are summed up to obtain the total importance index, and the names of the key target objects whose total importance index is greater than the importance index threshold are put into the subject category set.

Finally, all the category names in the subject category set are compared with all the category names in each subject set in the subject library to determine whether the category names in the subject set are the same as the category names in the subject category set. If they are the same, mark the category names in the corresponding subject set. After all the subject sets in the subject library have completed the comparison and judgment, rank the subject sets in ascending order according to the number of marked category names, and select the subject set ranked first as the current subject set.

The risk identification module includes a relevance judgment unit and a risk judgment unit.

The relevance judgment unit is used to judge the relevance of each target object with the video theme. The category name of the target object is compared with all category names in the current theme set to determine whether the same category name exists in the current theme set. If so, the relevance of the corresponding target object is set to strong relevance. If not, it continues to be compared with all category names in other theme sets in the theme library except the current theme set to determine whether the same category name exists. If so, the relevance of the corresponding target object is set to weak relevance. If not, the relevance of the corresponding target object is set to no relevance.

The risk judgment unit is used to calculate the risk index of each target object. Different influence coefficients are set for different associations. The influence coefficients are set in the order of no association, weak association and strong association from large to small. The influence coefficient and the number of pixels corresponding to the association degree of the target object are substituted into the formula to calculate the risk index.

A high correlation coefficient means that the target object is not closely related to the subject of the video, and the probability of a violation is higher. A larger number of pixels means that the target object's image accounts for a higher proportion of the entire image, and the impact of a violation is more severe. Comprehensively considering the correlation coefficient and the number of pixels can better assess the possible risks of the target object.

The operation management module is used to supervise risk objects. First, the number of callable devices in the distributed devices is obtained, and the target objects are ranked in positive order according to the risk index; second, the top I target objects are selected as risk objects, and devices are assigned to these risk objects in turn. The assigned devices monitor the areas where the risk objects are located; finally, when the risk objects are monitored as illegal items or illegal behaviors, pixel-level image processing is performed on the areas where the risk objects are located without affecting the overall clarity of the single-frame image.

Each device only monitors one risk object, and multiple devices can monitor and process multiple risk objects in a single frame image.

When the system is running, real-time video is continuously decomposed into single-frame images. After key area division and video subject identification, risk objects are quickly identified and distributed technology is used to use multiple devices to supervise the areas where different risk objects are located.

When some risk objects violate regulations, the computing power resources of multiple devices are used to perform fast image processing at the pixel level. The image processing is only for the area where the risk objects are located. After the processing is completed, the image of the processed area is quickly combined into the original single-frame image.

Embodiment 1:

Assume that the focal length of the live broadcast device camera at the corresponding time of a single-frame image is 1.5mm, the shooting distance is 2m, the focal length influence coefficient is 0.4, the maximum focal length of the live broadcast device camera is 3mm, the shooting distance influence coefficient is 0.2, and the distance constant is 10. Substitute it into the formula to calculate the weight ratio of the single-frame image:

Weight ratio:

Embodiment 2:

Assume that the position coordinates of a key target object are (250,500) and the number of pixels is 1200; the corresponding single-frame image center point position coordinates are (500,500) and the number of pixels in the key area is 25000; the distance influence coefficient is 0.0001 and the position constant is 1. Substitute the formula to calculate the importance index of the key target object:

Important index:

It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims (2)

1. The distributed video processing supervision method based on information security is characterized by comprising the following steps of: the method comprises the following steps:

s1, acquiring operation parameters of live broadcast equipment and shot real-time video in real time, and acquiring a latest theme library;

S2, dividing regions of the video picture by analyzing the operation parameters, identifying target objects and calculating important indexes;

S3, selecting a subject object according to the importance index, and judging the association degree of each target object;

S4, monitoring and image processing are carried out on target objects with different degrees of association by adopting a distributed technology;

In S1, the operation parameters refer to the focal length and shooting distance of a camera of the live broadcast equipment, the shooting distance refers to the distance between the camera and a shooting target, and a distance sensor arranged inside the live broadcast equipment is used for data acquisition; the subject library is used as a comparison library of live broadcast types and is used for identifying the type of each live broadcast, the subject library contains subject sets of various live broadcast types, and the subject sets contain category names of different objects under the same subject;

In S2, the specific steps are as follows:

S201, acquiring a real-time video shot by live broadcast equipment, decomposing the video into single-frame images by using the existing OpenCV, screening all the single-frame images, and removing the blurred single-frame images by adopting an edge detection method;

s202, analyzing the rest single-frame images one by one, obtaining the focal length and shooting distance of a live broadcast equipment camera under the corresponding time of each Shan Zhen image, substituting the focal length and shooting distance into a formula, and respectively calculating the weight ratio R of each Shan Zhen image; taking a single frame image center point as a center, taking the product of the single frame image length and the weight ratio R as a length, taking the product of the single frame image width and the weight ratio R as a width, and dividing a rectangular area as a key area; the weight ratio calculation formula is as follows:

Wherein R is weight duty ratio, alpha is focal length influence coefficient, J is focal length, J _max is equipment maximum focal length, beta is shooting distance influence coefficient, a is distance constant, and P is shooting distance;

S203, detecting all target objects in each Shan Zhen image by adopting a YOLOv target detection algorithm, giving identifiers, extracting characteristic information of the target objects, analyzing the type and the position of the target objects, calculating the number of pixels occupied by the target objects, marking the target objects with positions in a key area as key points, and marking other target objects as common;

S204, acquiring a position coordinate (x _i,y_i) of a key target object and a center point position coordinate (x _z,y_z) of a corresponding single-frame image, substituting the position coordinate and the pixel number of the key target object into a formula to calculate an important index, wherein each key target object corresponds to one important index; the calculation formula is as follows:

Wherein ZY is an important index, S _i is the number of pixels of a key target object, S _z is the number of pixels of a key region of a single-frame image, u is a distance influence coefficient, and c is a position constant;

S205, carrying out similarity recognition between target objects in different single-frame images, associating target objects which are higher than a similarity threshold and do not belong to the same Shan Zhen image, summing important indexes of the associated key target objects to obtain a total important index, reflecting the associated target objects in different single-frame images as the same target object, and unifying identifiers of all the associated target objects;

In S3, the specific steps are as follows:

S301, taking a key target object with a total importance index larger than an importance index threshold as a subject object, and putting names of the categories to which all the subject objects belong into a subject category set, wherein the set comprises { Q ₁,Q₂,Q₃,...,Q_n }, n represents the number of category names, and Q _n represents the nth category name;

S302, comparing all kinds of names in the topic class sets with all kinds of names in each topic set in the topic library respectively, judging whether the kinds of names in the topic sets are the same as the kinds of names in the topic class sets, marking the kinds of names in the corresponding topic sets if the kinds of names in the topic sets are the same, and after all topic sets in the topic library are subjected to comparison judgment, carrying out positive ranking on the topic sets according to the number of marked kinds of names, and selecting the topic set with the first ranking as the current topic set;

S303, obtaining the category names of all the target objects, respectively comparing the category names with all the category names in the current theme set, judging whether the same category names exist in the current theme set, setting the association degree of the corresponding target objects as strong association if the same category names exist, continuously comparing the category names with all the category names in other theme sets except the current theme set in the theme library if the same category names exist, judging whether the same category names exist, setting the association degree of the corresponding target objects as weak association if the same category names exist, and setting the association degree of the corresponding target objects as no association if the same category names do not exist;

in S4, the specific steps are as follows:

S401, setting different influence coefficients for different association degrees, wherein the influence coefficients are set from large to small according to the sequence of no association, weak association and strong association; acquiring the relevance of all the target objects, substituting the influence coefficient and the pixel number corresponding to the relevance of the target objects into a formula, and calculating a risk index; the formula is as follows:

FX＝K×E×S

Wherein FX is a risk index of the target object, K is an influence coefficient, E is a constant, and S is the number of pixels of the target object;

S402, acquiring the number I of callable devices in the distributed devices, carrying out positive sequence ranking on target objects according to risk indexes, selecting the target objects with the top I name as risk objects, sequentially distributing devices to the risk objects, and monitoring the distributed devices aiming at the areas where the risk objects are located;

S403, when the risk object is an illegal object or illegal behaviors are monitored, carrying out pixel-level image processing on the area where the risk object is located on the premise that the overall definition of the single-frame image is not affected, wherein the processing modes comprise mosaic pixelation, a pixel repairing method and a pixel replacing method.

2. The utility model provides a distributed video processing supervisory systems based on information security which characterized in that: the system comprises a data acquisition module, a data analysis module, a risk identification module and an operation management module;

The data acquisition module is used for acquiring the operation parameters of the live broadcast equipment and the shot real-time video; the data analysis module analyzes the video information frame by frame, divides key areas by combining the operation parameters of the equipment, identifies target objects, calculates important indexes of the target objects, and determines video subjects according to the important indexes; the risk identification module is used for judging the association degree of the target object and the video theme, substituting the association degree into a formula to calculate the risk index of each target object, and finding out the risk object to be monitored according to the risk index; the operation management module adopts a distributed technology to control different devices to monitor the area where each risk object is located, and when the risk object is an illegal object or has illegal behaviors, the image processing is carried out to protect the information security;

the data acquisition module comprises an image information acquisition unit, an equipment information acquisition unit and a theme information acquisition unit;

The image information acquisition unit is used for acquiring real-time video shot by live broadcast equipment; the equipment information acquisition unit is used for acquiring operation parameters of live equipment during working, wherein the operation parameters comprise a focal length and a shooting distance, and the shooting distance refers to the distance between the live equipment and a shooting target; the topic information acquisition unit is used for acquiring a topic library in the system, wherein the topic library comprises topic sets of various live broadcast types, and the topic sets comprise category names of different objects under the same topic;

The data analysis module comprises a region dividing unit, an object recognition unit and a theme analysis unit;

The region dividing unit is used for dividing key regions in the video picture; firstly, decomposing a real-time video shot by live broadcast equipment into a single-frame image; secondly, acquiring operation parameters of the live broadcast equipment under the corresponding time of each Shan Zhen image, and substituting the operation parameters into a formula to respectively calculate the weight ratio of each Shan Zhen image; finally, taking the center point of the single frame image as the center, taking the product of the length of the single frame image and the weight ratio R as the length, taking the product of the width of the single frame image and the weight ratio R as the width, dividing a rectangular area as a key area, and each Shan Zhen image is provided with a key area;

The weight ratio calculation formula is: Wherein alpha is a focal length influence coefficient, J is a focal length, J _max is a maximum focal length of the device, beta is a shooting distance influence coefficient, a is a distance constant, and P is a shooting distance;

The object recognition unit is used for recognizing key target objects; detecting target objects in each Shan Zhen image by adopting YOLOv target detection algorithm, extracting characteristic information of the target objects, analyzing the type and the position of the target objects, calculating the number of pixels occupied by the target objects, and taking the positions in a key area as key target objects;

The theme analysis unit is used for analyzing the video theme;

Firstly, acquiring position coordinates of an important target object and position coordinates of a central point of a corresponding single-frame image, substituting the position coordinates and the pixel number of the important target object into a formula to calculate an important index;

The calculation formula of the important index is as follows: Wherein S _i is the number of pixels of the key target object, S _z is the number of pixels of the key region of the single-frame image, (x _i,y_i) is the position coordinate of the key target object, (x _z,y_z) is the position coordinate of the central point of the single-frame image, u is the distance influence coefficient, and c is the position constant;

Secondly, correlating the same target object in different single-frame images, summing important indexes of correlated key target objects to obtain a total important index, and placing names of key target objects with the total important index being greater than an important index threshold value into a theme type set;

Finally, comparing all kinds of names in the topic class sets with all kinds of names in each topic set in the topic library respectively, judging whether the kinds of names in the topic sets are the same as the kinds of names in the topic class sets, marking the kinds of names in the corresponding topic sets if the kinds of names in the topic sets are the same, and after all topic sets in the topic library are subjected to comparison judgment, ranking the topic sets in positive order according to the number of marked kinds of names, and selecting the topic set with the first ranking as the current topic set;

The risk identification module comprises a relevance judgment unit and a risk judgment unit;

The relevance judging unit is used for judging the relevance of each target object and the video theme; comparing the category names of the target objects with all category names in the current theme set, judging whether the same category names exist in the current theme set, setting the association degree of the corresponding target objects as strong association if the same category names exist, continuously comparing the category names with all category names in other theme sets except the current theme set in the theme library if the same category names exist, judging whether the same category names exist, setting the association degree of the corresponding target objects as weak association if the same category names exist, and setting the association degree of the corresponding target objects as no association if the same category names do not exist;

the risk judging unit is used for calculating a risk index of each target object; setting different influence coefficients for different association degrees, setting the influence coefficients from large to small according to the sequence of no association, weak association and strong association, substituting the influence coefficients and the pixel number corresponding to the association degree of the target object into a formula, and calculating a risk index;

the risk index calculation formula is: k is an influence coefficient, E is a constant, and S is the number of pixels of the target object;

The operation management module is used for supervising the risk objects; firstly, acquiring the number I of callable devices in distributed devices, and carrying out positive ranking on target objects according to risk indexes; secondly, selecting the target objects with the top I names as risk objects, sequentially allocating equipment to the risk objects, and monitoring the area where the allocated equipment is located by aiming at the risk objects; and finally, when the risk object is monitored to be an illegal object or illegal behaviors occur, carrying out pixel-level image processing on the area where the risk object is located on the premise that the overall definition of the single-frame image is not affected.