A video production control method and system based on a cloud platform

Abstract

This invention relates to the field of control technology, and specifically discloses a video production control method and system based on a cloud platform. The method includes: real-time acquisition of multi-source scene data within a target scene; dynamic processing of the original video stream using preset edge processing rules to generate an intermediate video with attached attribute tags; determination of the scene action category of the intermediate video based on the parsing of the attribute tags, considering feature descriptions and application preferences; and determination of the action priority level of the scene action category based on in-depth analysis of user group video operations; matching the corresponding video production template according to the scene action category of the intermediate video; and dynamically allocating resources from the cloud platform resource pool according to the action priority level to personalize the intermediate video production. This enables intelligent control of video production, effectively improving the efficiency and quality of video production, and meeting the diverse needs of complex scenes and users.

Description

Technical Field

[0001] This invention relates to the field of control technology, and in particular to a video production control method and system based on a cloud platform. Background Technology

[0002] In today's digital age, the demand for video content is exploding and is widely used in numerous fields. Traditional video production methods often rely on local equipment and manual operation, resulting in low production efficiency and high costs.

[0003] With the development of cloud platform technology, although it has provided certain resource storage and processing capabilities for video production, existing cloud-based video production methods still have shortcomings such as a lack of targeted and personalized video production, and difficulty in coping with complex and ever-changing scene challenges. Therefore, how to achieve personalized video production and meet the diverse needs of complex scenarios and users has become one of the current research focuses.

[0004] Therefore, the present invention provides a video production control method and system based on a cloud platform. Summary of the Invention

[0005] This invention provides a cloud-based video production control method and system. It utilizes preset edge processing rules to dynamically process multi-source scene data, generating intermediate videos with attached attribute tags. Based on the parsing of the attribute tags, it comprehensively considers feature descriptions and application preferences to determine the scene action categories of the intermediate videos, and deeply analyzes user group video operations to determine the action priority levels of these categories. According to the scene action categories of the intermediate videos, it matches corresponding video production templates and dynamically allocates resources from the cloud platform resource pool according to action priority levels for personalized production of the intermediate videos. This enables intelligent control of video production, effectively improving production efficiency and quality, and meeting the diverse needs of complex scenarios and users.

[0006] This invention provides a video production control method based on a cloud platform, comprising:

[0007] Step 1: Collect multi-source scene data in the target scene in real time, and combine it with preset edge processing rules to dynamically process the original video stream and generate an intermediate video with attribute tags.

[0008] Step 2: Parse and analyze the attribute tag content to determine the scene action category and action priority level of the corresponding intermediate video;

[0009] Step 3: Based on the scene action category of the intermediate video, match the corresponding video production template, and then dynamically allocate resources from the cloud platform resource pool according to the action priority level to produce the intermediate video in a targeted manner.

[0010] Preferably, multi-source scene data within the target scene is acquired in real time, and combined with preset edge processing rules, the original video stream is dynamically processed to generate an intermediate video with attached attribute tags, including:

[0011] The system uses a pre-set acquisition terminal to collect various on-site data in the target scene in real time and transmits it to the pre-set edge processing node in real time.

[0012] The preset edge processing node matches the initial processing rules according to the data category, performs initial preprocessing on the received data, and then divides it into non-video data or video data.

[0013] The time base of non-video data and video data is synchronized using a preset buffering mechanism to generate an intermediate video file with attribute tags.

[0014] Preferably, based on the parsing of attribute tags, the scene action category of the intermediate video is determined by comprehensively considering feature descriptions and application preferences, and the action priority level of the scene action category is determined by in-depth analysis of the video operations of user groups, including:

[0015] The preset parsing algorithm is used to parse and transform the attribute tags attached to the intermediate video to obtain pre-analyzed information;

[0016] Feature extraction is performed on the pre-analysis information to obtain key analytical features;

[0017] The key analytical features are mapped to the preset action feature classes of the target scene to obtain action feature classes whose feature overlap exceeds a set overlap threshold, and these are regarded as reference feature classes.

[0018] When multiple reference feature classes exist, the actual feature class is determined by comprehensively considering the descriptive similarity between the feature classes and application preferences.

[0019] If a single reference feature class exists, it is output as the actual feature class.

[0020] By inputting key analytical features into an action analysis model that is adapted to the actual feature class, the scene action category of the corresponding intermediate video is obtained.

[0021] Using priority evaluation metrics that match the scene action categories, priority analysis is performed based on the parsed attribute label data to obtain action priority scores;

[0022] By utilizing the feature importance weights of the actual feature class to which the current scene action category belongs, the action priority score is optimized to obtain the priority evaluation coefficient;

[0023] Based on the priority evaluation coefficient, determine the action priority level of the intermediate video to which the action category of the current scene belongs.

[0024] Preferably, when multiple reference feature classes exist, the actual feature class is determined by comprehensively considering the descriptive similarity between feature classes and application preferences, including:

[0025] Extract the pre-defined action descriptions for each reference feature class and perform pairwise similarity comparisons to obtain the description similarity.

[0026] If no description similarity exceeds the set similarity threshold, the corresponding reference feature class with the highest feature overlap will be regarded as the actual feature class.

[0027] If the description similarity between two reference feature classes exceeds the set similarity threshold, and the description similarity is the maximum value among all similarity comparison results, then the two reference feature classes will be labeled as the first feature class and the second feature class, respectively.

[0028] The first feature class is identified as the feature with overlapping features through feature term mapping, and marked as the first analysis feature;

[0029] The second feature class is identified as the feature with overlapping features through feature term mapping and marked as the second analysis feature;

[0030] Based on the first or second analytical feature, the historical frequency of occurrence, video usage rate, and historical video rendering level within a preset time period, the overlap importance score of the first feature class or the second feature class is determined respectively.

[0031] The corresponding overlapping importance score of the first feature class or the second feature class is combined with the feature importance weight to calculate the corresponding comprehensive importance score;

[0032] The feature class with the highest overall importance score among the first and second feature classes is output as the actual feature class.

[0033] Preferred options also include:

[0034] Users are divided into user groups according to preset classification criteria, and each user group is labeled with a group category.

[0035] Regularly acquire historical video operation data for each user group, targeting various action characteristics;

[0036] Based on historical video operation data, determine the historical operation categories contained in each historical video of the action feature class and the historical execution count of the corresponding historical operation category;

[0037] By combining the preset attention representation weights of each historical operation category contained in the historical video with the corresponding historical execution count, a reference attention score for the current historical video is obtained.

[0038] For each historical video in the action feature category, a reference attention curve is constructed sequentially based on the historical video timestamps, using the reference attention score.

[0039] Perform trend analysis on the reference attention curve to determine the attention change coefficient of the current action feature class;

[0040] By comprehensively considering the user proportion of each user group and the corresponding attention change coefficient, the attention-adjustment coefficient of the corresponding action feature category is determined;

[0041] Based on the attention-adjustment coefficient, the feature importance weights of the action feature class are initially adjusted to obtain the pre-importance weights;

[0042] Sort the action feature classes according to their pre-importance weights from largest to smallest, and generate a list of undetermined feature classes;

[0043] Based on the scene action stage of the target scene, a corresponding pre-set important list of feature classes is matched and compared with the list of undetermined feature classes. Based on the ranking comparison result, the feature importance weight of each action feature class is determined.

[0044] Preferably, based on the ranking comparison results, the feature importance weight of each action feature class is determined, including:

[0045] If the list of undetermined feature classes is consistent with the list of pre-defined feature class importance, then the pre-importance weight of the action feature class will be output as the feature importance weight.

[0046] If the list of pending feature classes and the list of important preset feature classes are inconsistent, then the action feature classes with inconsistent sorting positions are regarded as adjustment feature classes, and the difference in sorting position of the adjustment feature classes is determined.

[0047] Based on the difference in sorting position and the preset adjustment unit, the pre-importance weights of the feature classes are adjusted and then output as the feature importance weights.

[0048] Preferably, based on the scene action categories of the intermediate video, a corresponding video production template is matched, and then resources are dynamically allocated from the cloud platform resource pool according to the action priority level to perform targeted production on the intermediate video, including:

[0049] Based on the scene action category of the current intermediate video, extract the video production template from the preset video production template library as the initial reference template;

[0050] The design rules for creating the initial reference template are compared with the actual design requirements to obtain the design comparison results;

[0051] Based on the design comparison results, if there are any non-compliant design elements, the design requirements for the non-compliant design elements are extracted from the actual design requirements, and the initial reference template is adjusted accordingly to obtain the actual reference template.

[0052] Based on the resources allocated from the cloud platform resource pool using action priority, and in accordance with the actual reference template, the intermediate video is processed to generate a video for pending requirements.

[0053] The pending demand videos are tested using quality inspection indicators, and when the quality inspection is qualified, they are stored as actual demand videos in the predetermined storage location of the cloud platform.

[0054] This invention provides a cloud-based video production and control system, comprising:

[0055] Data processing module: Used to collect multi-source scene data in the target scene in real time, and combine it with preset edge processing rules to dynamically process the original video stream and generate intermediate video with attribute tags;

[0056] Action Analysis Module: Based on the parsing of attribute tags, it comprehensively considers feature descriptions and application preferences to determine the scene action category of intermediate videos, and deeply analyzes the video operations of user groups to determine the action priority of scene action categories;

[0057] Video production module: Based on the scene and action categories of the intermediate video, it matches the corresponding video production template, and then dynamically allocates resources from the cloud platform resource pool according to the action priority level to produce the intermediate video in a targeted manner.

[0058] The beneficial effects of this invention compared to existing technologies are as follows: By utilizing preset edge processing rules to dynamically process multi-source scene data, intermediate videos with attached attribute tags are generated; based on the parsing of attribute tags, the scene action category of the intermediate video is determined by comprehensively considering feature descriptions and application preferences, and the action priority level of the scene action category is determined by in-depth analysis of the video operations of user groups; according to the scene action category of the intermediate video, corresponding video production templates are matched, and resources are dynamically allocated from the cloud platform resource pool according to the action priority level to perform personalized production of the intermediate video. This enables intelligent control of video production, effectively improves the efficiency and quality of video production, and meets the diverse needs of complex scenes and users.

[0059] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0060] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0061] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0062] Figure 1 This is a schematic diagram of a video production control method based on a cloud platform in an embodiment of the present invention;

[0063] Figure 2 This is a structural diagram of a cloud-based video production and control system according to an embodiment of the present invention. Detailed Implementation

[0064] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention. Example 1:

[0065] This invention provides a video production control method based on a cloud platform, with reference to... Figure 1 ,include:

[0066] Step 1: Collect multi-source scene data in the target scene in real time, and combine it with preset edge processing rules to dynamically process the original video stream and generate an intermediate video with attribute tags.

[0067] Step 2: Based on the parsing of attribute tags, the scene action category of the intermediate video is determined by comprehensively considering feature descriptions and application preferences, and the action priority of the scene action category is determined by in-depth analysis of the video operations of the user group.

[0068] Step 3: Based on the scene action category of the intermediate video, match the corresponding video production template, and then dynamically allocate resources from the cloud platform resource pool according to the action priority level to produce the intermediate video in a targeted manner.

[0069] In this embodiment, the target scene refers to the specific environment in which video production is required, such as an industrial production workshop; multi-source scene data refers to all types of information related to production, equipment operation, and environmental status from the target scene, including but not limited to video data, audio data, sensor data (such as temperature, humidity, pressure, etc.), and equipment status data.

[0070] In this embodiment, the preset edge processing rules refer to the pre-defined rules for processing the acquired data at the edge processing nodes. The edge processing nodes are usually located near the data acquisition end and are used to process data quickly and reduce data transmission latency. The raw video stream refers to the unprocessed video data stream directly obtained from the acquisition device.

[0071] In this embodiment, attribute tags refer to the identification information attached to the intermediate video, which is used to describe the relevant attributes of the video, such as scene type, action category, timestamp, data source, etc.; intermediate video refers to the video file with attribute tags generated after the original video stream is dynamically processed by the edge processing node.

[0072] In this embodiment, the scene action category refers to the specific action category determined for the target scene of a specific application; the action priority level is used to indicate the priority of scene actions in the intermediate video; the video production template is a set of video production styles and rules pre-set for different scene action categories, which is used to produce intermediate videos in a targeted manner. For example, for the scene action category of equipment failure, the video production template may include highlighting the faulty equipment and adding fault description text; for the production process category, it may include showing the process flow and close-ups of key steps; the cloud platform resource pool refers to the collection of various computing, storage, network and other resources stored and managed on the cloud platform. When producing intermediate videos, resources are dynamically allocated from this resource pool according to the action priority level.

[0073] The beneficial effects of the above technologies are as follows: By utilizing preset edge processing rules to dynamically process multi-source scene data, intermediate videos with attached attribute tags are generated; based on the parsing of attribute tags, the scene action category of the intermediate video is determined by comprehensively considering feature descriptions and application preferences, and the action priority level of the scene action category is determined by in-depth analysis of the video operations of user groups; according to the scene action category of the intermediate video, the corresponding video production template is matched, and then resources are dynamically allocated from the cloud platform resource pool according to the action priority level to perform personalized production of the intermediate video. This enables intelligent control of video production, effectively improves the efficiency and quality of video production, and meets the diverse needs of complex scenarios and users. Example 2:

[0074] This invention provides a cloud-based video production control method that collects multi-source scene data within a target scene in real time and dynamically processes the original video stream using preset edge processing rules to generate an intermediate video with attached attribute tags, including:

[0075] The system uses a pre-set acquisition terminal to collect various on-site data in the target scene in real time and transmits it to the pre-set edge processing node in real time.

[0076] The preset edge processing node matches the initial processing rules according to the data category, performs initial preprocessing on the received data, and then divides it into non-video data or video data.

[0077] The time base of non-video data and video data is synchronized using a preset buffering mechanism to generate an intermediate video file with attribute tags.

[0078] In this embodiment, the preset acquisition terminal refers to a set of devices pre-set for acquiring various data in the target scene, such as temperature sensors, pressure sensors, and cameras; the target scene refers to the specific application scenario targeted by the video production control method, such as a factory production workshop, which includes various production equipment, material flow, and personnel operations; or it can be a logistics warehouse, involving the storage, handling, and sorting of goods.

[0079] In this embodiment, "on-site data" refers to all types of information generated within the target scene that are related to production, equipment operation, and environmental status. Examples include video data (such as images captured by cameras), audio data (such as the sounds of equipment operation), sensor data (such as measurements of physical quantities like temperature, humidity, pressure, and displacement), and equipment status data (such as the on / off status and operating speed of equipment). Pre-deployed edge processing nodes refer to processing units pre-deployed at the network edge near the data acquisition source. These nodes possess certain computing and storage capabilities, enabling them to perform preliminary processing and analysis on the acquired data before transmitting it to the cloud platform, reducing the time and bandwidth consumption of data transmission to the cloud platform. Examples include local servers and smart gateway devices within the target scene. Data categories refer to classifying the acquired data according to its characteristics, including video data, sensor data, and audio data.

[0080] In this embodiment, the initial processing rules refer to the processing methods and processes pre-defined for different data categories. For example, for video data, the initial processing rules include image compression, noise reduction, and target detection; for sensor data, they include data calibration, filtering, and outlier detection; and for audio data, they include audio noise reduction, audio feature extraction, and audio segmentation. The preset buffer mechanism is used to temporarily store data using a buffer to resolve the temporal inconsistency of different data types and ensure that they can be processed and analyzed synchronously. This usually refers to a double buffer mechanism. The time base is used to measure and synchronize the time standard of different data.

[0081] In this embodiment, attribute tags are used to describe the identification of various data features and information contained in the intermediate video file, such as the specific time of video shooting, shooting location (e.g., specific workshop location or warehouse area), camera number, name of the main target object detected in the video (e.g., the product model being processed on the production line), key physical quantity values ​​collected by sensors (e.g., temperature value, pressure value at a certain moment), equipment status indicators (e.g., whether the equipment is in normal operation, fault shutdown, or standby state), etc. The intermediate video file refers to the file generated after initial preprocessing by the edge processing node and synchronization of the time base of non-video data and video data using a buffer mechanism. The file not only contains video data but also comes with relevant attribute tags, providing rich information for subsequent video production control and analysis.

[0082] The beneficial effects of the above technologies are as follows: by using pre-set edge processing nodes to preprocess and classify various in-scene data in real-time acquisition of the target scene, and then using a preset buffer mechanism to synchronize the time base of non-video data and video data, an intermediate video file with attribute tags is generated. This can effectively reduce the time and bandwidth consumption of data transmission to the cloud, improve data processing efficiency, and provide rich information for subsequent video production control and analysis. Example 3:

[0083] This invention provides a cloud-based video production control method. Based on the parsing of attribute tags, it comprehensively considers feature descriptions and application preferences to determine the scene action categories of intermediate videos, and deeply analyzes user group video operations to determine the action priority level of scene action categories, including:

[0084] The preset parsing algorithm is used to parse and transform the attribute tags attached to the intermediate video to obtain pre-analyzed information;

[0085] Feature extraction is performed on the pre-analysis information to obtain key analytical features;

[0086] The key analytical features are mapped to the preset action feature classes of the target scene to obtain action feature classes whose feature overlap exceeds a set overlap threshold, and these are regarded as reference feature classes.

[0087] When multiple reference feature classes exist, the actual feature class is determined by comprehensively considering the descriptive similarity between the feature classes and application preferences.

[0088] If a single reference feature class exists, it is output as the actual feature class.

[0089] By inputting key analytical features into an action analysis model that is adapted to the actual feature class, the scene action category of the corresponding intermediate video is obtained.

[0090] Using priority evaluation metrics that match the scene action categories, priority analysis is performed based on the parsed attribute label data to obtain action priority scores;

[0091] By utilizing the feature importance weights of the actual feature class to which the current scene action category belongs, the priority score of the corresponding action is optimized to obtain the priority evaluation coefficient;

[0092] Based on the priority evaluation coefficient, determine the action priority level of the intermediate video to which the action category of the current scene belongs.

[0093] In this embodiment, the preset parsing algorithm refers to a pre-set algorithm used to parse the content and convert the data of the attribute tags attached to the intermediate video. It can convert the complex and unstructured information in the attribute tags into a format that computers can understand and process, such as a regular expression matching algorithm. The pre-analyzed information refers to the information obtained after the preset parsing algorithm parses the content and converts the data of the attribute tags attached to the intermediate video, such as the shooting time and shooting location.

[0094] In this embodiment, key analytical features refer to features extracted from pre-analyzed information using specific feature extraction algorithms (usually based on word frequency statistics, semantic analysis, etc.) that can represent the core characteristics of the pre-analyzed information. For example, if the pre-analyzed information includes the video shooting location as "Production Line 1 of Workshop A in Factory A" and the shooting object as "Product X", then the key analytical features are "Factory A", "Production Line 1" and "Product X", which can clearly identify the scene and object of the video.

[0095] In this embodiment, the action feature class refers to a set of pre-defined action feature categories for a specific application scenario, including equipment failure categories (such as machine shutdown, abnormal noise), production process categories (such as product assembly, parts processing), equipment teaching categories (such as training new employees to operate equipment), etc.

[0096] In this embodiment, feature mapping refers to the process of comparing key analysis features with feature items in the preset action feature class of the target scene. For example, the feature items with key analysis features include "machine shutdown" and "abnormal sound". The preset action feature class of the target scene includes a device failure class, which contains the feature item "machine shutdown". Through feature mapping, it can be found that the key analysis features and the device failure class have the same feature item "machine shutdown".

[0097] In this embodiment, the overlap threshold is a pre-set quantitative value used to determine whether the overlap between key analysis features and action feature classes reaches a certain standard, such as 0.6. The feature overlap is quantified by dividing the number of overlaps between the key analysis features and the current action feature class by the total number of all feature items in the corresponding action feature class. The reference feature class refers to the action feature class whose feature overlap exceeds the set overlap threshold.

[0098] In this embodiment, description similarity is used to measure the degree of similarity between reference feature classes in preset action descriptions, where preset action descriptions refer to text information that provides a detailed description of the action feature class; description similarity is usually obtained using the cosine similarity algorithm.

[0099] In this embodiment, application preference is used to reflect the degree of attention a user pays to different action feature classes; actual feature class refers to the feature class determined by comprehensively considering the descriptive similarity between feature classes and application preference when there are multiple reference feature classes, or the corresponding reference feature class when there is a single reference feature class.

[0100] In this embodiment, the action analysis model is an analysis model based on actual feature class matching. Different action feature classes correspond to different action analysis models. The action analysis model is obtained by extracting features from relevant data of the action feature class and then training a neural network. It can accurately determine the action category based on the input features. For example, when the action feature class is equipment failure, the corresponding action analysis model is a neural network model trained on a large amount of equipment failure data. Inputting key analysis features such as machine downtime and abnormal noise frequency into this model, the model can output the specific equipment failure type, such as motor failure or transmission system failure, i.e., the scene action category. When the action feature class is production process, the corresponding action analysis model is a neural network model trained on a large amount of production process-related data. This model can accurately determine the specific production process category based on the input key analysis features, such as process steps and production equipment status, such as parts assembly process or product inspection process. When the action feature class is equipment teaching, the corresponding action analysis model is a neural network model trained on a large amount of equipment teaching-related data. This model can accurately determine the specific equipment teaching content category based on the input key analysis features, such as teaching operation steps and equipment component explanations, such as equipment startup teaching or equipment maintenance teaching.

[0101] In this embodiment, the priority evaluation index refers to a pre-set index that matches the scene action category and is used to prioritize the analysis of the parsed attribute tag data. Different scene action categories correspond to different priority evaluation indices. For example, for the scene action category of equipment failure, the priority evaluation indices include the severity of the failure (such as whether it causes a complete production stoppage), the scope of impact (such as the number of production lines affected), and the difficulty of repair. For the scene action category of production process, the priority evaluation indices include the importance of the process (such as whether it is a critical process) and the impact on the production progress (such as the number of productions affected).

[0102] In this embodiment, the action priority score is a weighted average of the indicator evaluation scores obtained by performing priority analysis on the parsed attribute label data using priority evaluation indicators that match the action category in the scene. The weights assigned to the priority evaluation indicators are obtained by solving the matrix constructed by pairwise comparison and scoring using the analytic hierarchy process, and all have values ​​of (0,1). The feature importance weight is used to reflect the importance of the actual feature class to which the current action category belongs, and all have values ​​of (0,1). The priority evaluation coefficient is a coefficient obtained by directly multiplying the feature importance weight of the actual feature class to which the current action category belongs by the action priority score, and is used to reflect the priority of the action.

[0103] In this embodiment, the action priority level refers to the level determined from the preset priority list based on the priority evaluation coefficient, which is used to represent the priority of scene actions in the intermediate video. The preset priority list consists of the value range of the priority evaluation coefficient and the corresponding priority level, which includes three levels: high, medium, and low. The preset priority list is established by reasonably dividing the value range of the priority evaluation coefficient based on actual business needs and historical data analysis, and then defining the corresponding priority level for each divided priority evaluation coefficient range.

[0104] The beneficial effects of the above technologies are: by determining the scene action categories in the intermediate video, and then combining the priority evaluation indicators and feature importance weights to determine the action priority level, it is possible to accurately identify scene actions in the video, and provide a reliable basis for the subsequent reasonable allocation of resources for video production based on the importance of actions, thereby improving the pertinence and efficiency of video production and meeting the diverse needs of video production in different scenarios. Example 4:

[0105] This invention provides a cloud-based video production control method. When multiple reference feature classes exist, the actual feature class is determined by comprehensively considering the descriptive similarity between feature classes and application preferences, including:

[0106] Extract the pre-defined action descriptions for each reference feature class and perform pairwise similarity comparisons to obtain the description similarity.

[0107] If no description similarity exceeds the set similarity threshold, the corresponding reference feature class with the highest feature overlap will be regarded as the actual feature class.

[0108] If the description similarity between two reference feature classes exceeds the set similarity threshold, and the description similarity is the maximum value among all similarity comparison results, then the two reference feature classes will be labeled as the first feature class and the second feature class, respectively.

[0109] The first feature class is identified as the feature with overlapping features through feature term mapping, and marked as the first analysis feature;

[0110] The second feature class is identified as the feature with overlapping features through feature term mapping and marked as the second analysis feature;

[0111] Based on the first or second analytical feature, the historical frequency of occurrence, video usage rate, and historical video rendering level within a preset time period, the overlap importance score of the first feature class or the second feature class is determined respectively.

[0112] The corresponding overlapping importance score of the first feature class or the second feature class is combined with the feature importance weight to calculate the corresponding comprehensive importance score;

[0113] The feature class with the highest overall importance score among the first and second feature classes is output as the actual feature class.

[0114] In this embodiment, the preset action description refers to text information that describes the action feature class in detail. For example, the equipment failure class is described as "the machine stops abnormally and is accompanied by abnormal noise"; the production process abnormal class is described as "the machine suddenly stops during the production process, affecting the production progress".

[0115] In this embodiment, the steps for obtaining the first feature class and the second feature class are as follows: when the description similarity between two reference feature classes exceeds the set similarity threshold, and the description similarity is the maximum value among all similarity comparison results, either of the two reference feature classes is labeled as the first feature class, and the other feature class is labeled as the second feature class, for subsequent further analysis to determine the actual feature class.

[0116] In this embodiment, for example, there is a reference feature class. , , After comparing the description similarity, it was found that and The description similarity between them is 0.85. and The description similarity between them is 0.8. and The description similarity between them is 0.5, and the similarity threshold is set to 0.7;

[0117] At this time, due to and The description similarity between them is 0.85, and... and The description similarity of 0.8 between all of them exceeds the set similarity threshold. Among all similarity comparison results, and The description similarity between them is the maximum value;

[0118] Ultimately, Labeled as the first feature class, It is labeled as the second feature class.

[0119] In this embodiment, the overlap importance score refers to a quantitative indicator of the importance of overlapping features within the first or second feature class, determined based on the historical frequency of occurrence of the first and second analytical features within a preset time period, video usage rate, and historical video rendering level. Historical frequency of occurrence refers to the number of times the first or second analytical feature has appeared in the preset time period; video usage rate refers to the proportion of videos containing the first or second analytical feature to the total number of videos within the preset time period; historical video rendering level refers to the rendering quality level set during the production process of videos containing the first or second analytical feature within the preset time period. Video rendering levels are divided into low, medium, and high, with different importance weights and rendering operations corresponding to different rendering levels. The importance weights are pre-set based on actual business needs and expert experience, with high rendering levels having the highest importance weights and low rendering levels having the lowest, all ranging from (0,1).

[0120] In this embodiment, the formula for calculating the overlapping importance score is as follows:

[0121] ;

[0122] In the formula, P represents the overlap importance score of the current first feature class or second feature class; v represents the historical occurrence frequency of the first analytical feature or second analytical feature corresponding to the current first feature class or second feature class. denoted as the weight of the historical frequency of occurrence on the importance score of the analysis overlap; c represents the video usage rate of the first or second analysis feature corresponding to the current first or second feature class. This is expressed as the weight of the impact of video usage rate on the analysis of overlapping importance scores; This represents the weight of the impact of historical rendering levels on the analysis of the importance score of overlap. It represents the number of historical videos of the i-th video rendering level in a video containing either the first or second analytical feature, where i = 1, 2, 3; This refers to the number of historical videos with a low video rendering level that contain either the first or second analytical feature. This refers to the number of historical videos at a video rendering level that contain either the first or second analytical feature. The first analysis feature is represented by the number of historical videos with a high video rendering level that contain either the first or second analysis feature; M represents the number of videos containing either the first or second analysis feature within a preset time period. Let represent the important representation weights for the i-th video rendering level, with values ​​all ranging from (0, 1); where, Represented as the weight of important representations at low video rendering levels. Represented as the important representation weight in the video rendering level. This represents the weight of important representations for high video rendering levels.

[0123] In this embodiment, a comprehensive importance score is used to measure the importance of a feature class, thereby determining the actual feature class; for example, there exists a first feature class. The overlap importance score is 0.6, and the feature importance weight is 0.7. Multiplying them, we get 0.6 × 0.7 = 0.42, resulting in a comprehensive importance score of 0.42. (Second feature class) The overlap importance score is 0.4, and the feature importance weight is also 0.7, resulting in a comprehensive importance score of 0.28.

[0124] At this time, due to the first feature class The first feature class has the highest overall importance score, so it is output as the actual feature class.

[0125] The beneficial effects of the above technologies are: by comprehensively considering factors such as the degree of similarity in description between feature classes, historical frequency of occurrence, video usage rate, and historical video rendering level, the actual feature classes can be accurately screened, and the action feature classes that best match the intermediate video can be accurately identified, thereby improving the accuracy of scene action category judgment and providing a more reliable foundation for subsequent video production and action priority evaluation. Example 5:

[0126] This invention provides a video production control method based on a cloud platform, which further includes:

[0127] Users are divided into user groups according to preset classification criteria, and each user group is labeled with a group category.

[0128] Regularly acquire historical video operation data for each user group, targeting various action characteristics;

[0129] Based on historical video operation data, determine the historical operation categories contained in each historical video of the action feature class and the historical execution count of the corresponding historical operation category;

[0130] By combining the preset attention representation weights of each historical operation category contained in the historical video with the corresponding historical execution count, a reference attention score for the current historical video is obtained.

[0131] For each historical video in the action feature category, a reference attention curve is constructed sequentially based on the historical video timestamps, using the reference attention score.

[0132] Perform trend analysis on the reference attention curve to determine the attention change coefficient of the current action feature class;

[0133] By comprehensively considering the user proportion of each user group and the corresponding attention change coefficient, the attention-adjustment coefficient of the corresponding action feature category is determined;

[0134] Based on the attention-adjustment coefficient, the feature importance weights of the action feature class are initially adjusted to obtain the pre-importance weights;

[0135] Sort the action feature classes according to their pre-importance weights from largest to smallest, and generate a list of undetermined feature classes;

[0136] Based on the scene action stage of the target scene, a corresponding pre-set important list of feature classes is matched and compared with the list of undetermined feature classes. Based on the ranking comparison result, the feature importance weight of each action feature class is determined.

[0137] In this embodiment, the preset classification standard refers to the pre-defined rules for classifying users in a scenario, which are usually based on the user's role (such as administrator, equipment operator, etc.); the user group refers to the set of users obtained after classifying users in a scenario according to the preset classification standard; the group category is used to identify the user group. For example, the user group obtained after classifying users in a scenario according to their role includes the administrator group, the equipment operator group, etc.

[0138] In this embodiment, historical video operation data refers to the operation records of each user group on videos of various action feature categories over a period of time, including operation categories (e.g., play, pause, fast forward, etc.) and operation time; historical operation category refers to the operation type obtained by classifying historical video operation data, such as play, pause, favorite, share, etc.; historical execution count refers to the number of times each historical operation category contained in each historical video has been executed over a period of time.

[0139] In this embodiment, the preset attention representation weight refers to the weight value set in advance based on expert evaluation and historical data analysis to measure the degree of influence of different historical operation categories on the user's attention level, and the value range is (0,1).

[0140] In this embodiment, the reference attention score is a score calculated by combining the preset attention representation weights of each historical operation category contained in the historical video with the corresponding historical execution counts. It is used to reflect the user's attention to the historical video. For example, there is a historical video 1 that includes three operation categories: play, favorite, and pause. The play operation weight is 0.6, the favorite operation weight is 0.3, and the pause operation weight is 0.1, with corresponding historical execution counts of 10, 2, and 3 times, respectively. In this case, the reference attention score = 0.6×10 + 0.3×2 + 0.1×3 = 6 + 0.6 + 0.3 = 6.9 points.

[0141] In this embodiment, the reference attention curve refers to the reference attention score for each historical video of the action feature category, which is constructed sequentially based on the timestamps of the historical videos. It is used to show the trend of user attention to action feature category videos over time. The specific steps for obtaining the attention change coefficient are as follows: first, the trend features (including slope, peak value, and volatility, etc.) of the reference attention curve are extracted; then, the extracted trend features are directly weighted and averaged to measure the change in the attention level of the current action feature category. The weights assigned to the trend features are determined by pairwise comparisons to determine the relative importance of each trend feature, and a judgment matrix is ​​constructed for calculation. The values ​​range from (0, 1).

[0142] In this embodiment, the attention-adjustment coefficient is a coefficient obtained by comprehensively considering the user proportion of each user group and the corresponding attention change coefficient, which is used to adjust the feature importance weight of the action feature class; the pre-importance weight is the weight value obtained after the initial adjustment of the feature importance weight of the action feature class based on the attention-adjustment coefficient.

[0143] In this embodiment, for example, there are three user groups in target scenario 1: administrators, equipment operators, and students. The administrator group accounts for 0.2% of users and has a change coefficient of 0.8 for attention to action feature category 1; the equipment operator group accounts for 0.3% of users and has a change coefficient of 0.7 for attention to action feature category 1; and the student group accounts for 0.5% of users and has a change coefficient of 0.9 for attention to action feature category 1.

[0144] At this time, action feature class 1 = 0.2×0.8+ 0.3×0.7+ 0.5×0.9 =0.82; Using the attention-adjustment coefficient of 0.82, the feature importance weight of action feature class 1 (0.5) is adjusted to obtain the pre-importance weight = ;in, It represents the sum of attention-adjustment coefficients for all action feature classes.

[0145] In this embodiment, the undetermined feature class list refers to the list generated after sorting the action feature classes from largest to smallest according to their pre-importance weights; the scene action stage refers to the specific action state of the target scene at different time nodes, such as the preparation stage, the output stage, and the maintenance stage; the pre-set feature class importance list refers to the list of action feature classes ranked by importance according to the scene action stages of the target scene. The importance ranking list of action feature classes is established for each scene action stage based on specific tasks, goals, and requirements, by using expert evaluation and the analytic hierarchy process to evaluate the importance of action feature classes. Specifically, it is constructed by sorting the action feature classes from largest to smallest according to their importance evaluation scores.

[0146] The beneficial effects of the above technologies are: by regularly acquiring historical video operation data of user groups, analyzing changes in users' attention to different action feature categories, and comprehensively considering user proportion and attention change coefficients to adjust the feature importance weight of action feature categories, it helps subsequent video production to better meet user expectations and improve the relevance of video production and user satisfaction. Example 6:

[0147] This invention provides a cloud-based video production control method, which determines the feature importance weight of each action feature class based on the ranking comparison results, including:

[0148] If the list of undetermined feature classes is consistent with the list of pre-defined feature class importance, then the pre-importance weight of the action feature class will be output as the feature importance weight.

[0149] If the list of pending feature classes and the list of important preset feature classes are inconsistent, then the action feature classes with inconsistent sorting positions are regarded as adjustment feature classes, and the difference in sorting position of the adjustment feature classes is determined.

[0150] Based on the difference in sorting position and the preset adjustment unit, the pre-importance weights of the feature classes are adjusted and then output as the feature importance weights.

[0151] In this embodiment, the adjustment feature class refers to the action feature class whose sorting position is inconsistent between the list of pending feature classes and the list of preset important feature classes; the sorting position difference refers to the difference between the position sorting of the adjustment feature class in the list of pending feature classes and the position sorting in the list of preset important feature classes.

[0152] In this embodiment, for example, if action feature class 2 is ranked 2 in the list of undetermined feature classes and ranked 3 in the list of preset important feature classes, then action feature class 2 is an adjustment feature class and the corresponding ranking position difference is 1.

[0153] If action feature class 3 is ranked 3rd in the list of undetermined feature classes and 2nd in the list of important preset feature classes, then action feature class 3 is an adjustment feature class, and the difference in the corresponding ranking position is -1.

[0154] In this embodiment, the preset adjustment unit refers to a pre-set unit value used to adjust the pre-importance weights of the adjustment feature classes. The acquisition steps specifically refer to: first, determining the maximum and minimum values ​​of the feature importance weights of all action feature classes before adjustment; then, averaging the obtained maximum and minimum values ​​of the feature importance weights to obtain the weight average; and finally, taking 10% of the weight average as the preset adjustment unit for the current adjustment.

[0155] In this embodiment, for example, if the pre-importance weight of feature class 1 is adjusted to 0.5, its position in the undetermined feature class list is 3, its position in the preset feature class importance list is 1, the difference in sorting position is -2, and the preset adjustment unit is 0.02, then the adjusted pre-importance weight of feature class 1 is = .

[0156] The beneficial effects of the above technology are: by sorting and comparing the list of undetermined feature classes with the list of preset feature class importance, and adjusting the feature importance weight of the action feature class according to the sorting and comparison results, the feature importance weight can be made more in line with the scene action stage requirements of the target scene, further optimizing resource allocation and action priority evaluation in the video production process, and improving the quality and efficiency of video production. Example 7:

[0157] This invention provides a cloud-based video production control method. Based on the scene action categories of the intermediate video, it matches corresponding video production templates, and then dynamically allocates resources from the cloud platform resource pool according to action priority levels to perform targeted production on the intermediate video. The method includes:

[0158] Based on the scene action category of the current intermediate video, extract the video production template from the preset video production template library as the initial reference template;

[0159] The design rules for creating the initial reference template are compared with the actual design requirements to obtain the design comparison results;

[0160] Based on the design comparison results, if there are any non-compliant design elements, the design requirements for the non-compliant design elements are extracted from the actual design requirements, and the initial reference template is adjusted accordingly to obtain the actual reference template.

[0161] Based on the resources allocated from the cloud platform resource pool using action priority, and in accordance with the actual reference template, the intermediate video is processed to generate a video for pending requirements.

[0162] The pending demand videos are tested using quality inspection indicators, and when the quality inspection is qualified, they are stored as actual demand videos in the predetermined storage location of the cloud platform.

[0163] In this embodiment, the video production template refers to a pre-designed video production framework and style, including screen layout, subtitle style, special effects settings, transition effects, etc., used to quickly generate videos that meet specific requirements; the preset video production template library refers to a database that stores various types of video production templates, from which users can select appropriate templates as needed; the initial reference template refers to a video production template extracted from the preset video production template library based on the scene action category of the intermediate video, serving as the initial reference for video production.

[0164] In this embodiment, the production design rules refer to the various design elements and production requirements specified in the video production template, such as font size, color matching, and screen ratio; the actual design requirements refer to the actual requirements of the user or business for video production, which may differ from the design rules of the initial reference template; the design comparison result refers to the result obtained by comparing the production design rules of the initial reference template with the actual design requirements, which is used to determine whether the template needs to be adjusted; the actual reference template refers to the video production template obtained after adjusting the initial reference template according to the actual design requirements.

[0165] In this embodiment, the cloud platform resource pool refers to the collection of various computing, storage, network and other resources integrated in the cloud platform to support various tasks in the video production process; pending requirements refer to the preliminary video files generated after the intermediate video is produced according to the actual reference template, without quality testing.

[0166] In this embodiment, the quality inspection index refers to a series of standards used to evaluate the quality of the pending demand video, including picture clarity, audio quality, and content integrity; the actual demand video refers to the pending demand video that has passed the quality inspection and meets the requirements of the user or business; the predetermined storage location refers to the location in the cloud platform that is pre-designated for storing the actual demand video, so that users can manage and access it.

[0167] The beneficial effects of the above technologies are: by accurately matching video templates based on scene action categories, flexibly adjusting according to actual needs, dynamically allocating resources for production according to action priority levels, and finally rigorously testing video quality, it is possible to ensure the efficient generation of high-quality videos that meet scene business requirements, thereby improving video production efficiency and quality. Example 8:

[0168] This invention provides a cloud-based video production and control system, with reference to... Figure 2 ,include:

[0169] Data processing module: Used to collect multi-source scene data in the target scene in real time, and combine it with preset edge processing rules to dynamically process the original video stream and generate intermediate video with attribute tags;

[0170] Action Analysis Module: Based on the parsing of attribute tags, it comprehensively considers feature descriptions and application preferences to determine the scene action category of intermediate videos, and deeply analyzes the video operations of user groups to determine the action priority of scene action categories;

[0171] Video production module: Based on the scene and action categories of the intermediate video, it matches the corresponding video production template, and then dynamically allocates resources from the cloud platform resource pool according to the action priority level to produce the intermediate video in a targeted manner.

[0172] The beneficial effects of the above technologies are as follows: By utilizing preset edge processing rules to dynamically process multi-source scene data, intermediate videos with attached attribute tags are generated; based on the parsing of attribute tags, the scene action category of the intermediate video is determined by comprehensively considering feature descriptions and application preferences, and the action priority level of the scene action category is determined by in-depth analysis of the video operations of user groups; according to the scene action category of the intermediate video, the corresponding video production template is matched, and then resources are dynamically allocated from the cloud platform resource pool according to the action priority level to perform personalized production of the intermediate video. This enables intelligent control of video production, effectively improves the efficiency and quality of video production, and meets the diverse needs of complex scenarios and users.

[0173] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A cloud platform-based video production control method, characterized by, The method comprises the following steps: Step 1: Real-time acquisition of multi-source scene data in the target scene, and dynamic processing of the original video stream combined with preset edge processing rules to generate intermediate videos with attribute tags; Step 2: Based on the analysis of the attribute tags, the scene action category of the intermediate video is determined by comprehensively considering the feature description and application preference, and the action priority level of the scene action category is determined by deep analysis of the video operation of the user group; Step 3: According to the scene action category of the intermediate video, the corresponding video production template is matched, and then the resources are dynamically allocated from the cloud platform resource pool according to the action priority level to produce the intermediate video; Wherein, based on the analysis of the attribute tags, the scene action category of the intermediate video is determined by comprehensively considering the feature description and application preference, and the action priority level of the scene action category is determined by deep analysis of the video operation of the user group, comprising: Using a preset analysis algorithm to perform content analysis and data conversion on the attribute tags attached to the intermediate video to obtain pre-analysis information; Feature extraction is performed on the pre-analysis information to obtain key analysis features; Map the key analysis features to the action feature classes pre-set for the target scene to obtain action feature classes with feature item coincidence degrees exceeding a set coincidence threshold, and regard them as reference feature classes; When there are multiple reference feature classes, the actual feature class is determined by comprehensively considering the description similarity and application preference between the reference feature classes; When there is only one reference feature class, it is output as the actual feature class; Input the key analysis features into the action analysis model adapted to the actual feature class to obtain the scene action category of the corresponding intermediate video; Using the priority evaluation index matched with the scene action category, perform priority analysis based on the analyzed attribute tag data to obtain an action priority score; Optimize the action priority score using the feature importance weight of the actual feature class to which the current scene action category belongs to obtain a priority evaluation coefficient; Determine the action priority level of the intermediate video to which the current scene action category belongs according to the priority evaluation coefficient. 2.The cloud platform-based video production control method of claim 1, wherein, Real-time acquisition of multi-source scene data in the target scene, and dynamic processing of the original video stream combined with preset edge processing rules to generate intermediate videos with attribute tags, comprising: Real-time acquisition of various in-scene data in the target scene using a preset acquisition terminal, and real-time transmission to a pre-set edge processing node; The pre-set edge processing node matches an initial processing rule according to the data category, and after initial preprocessing of the received data, divides the data into non-video data or video data; Synchronize the time base of non-video data and video data using a preset buffering mechanism to generate intermediate video files with attribute tags. 3.The cloud platform-based video production control method of claim 1, wherein, When there are multiple reference feature classes, the actual feature class is determined by comprehensively considering the description similarity and application preference between the feature classes, comprising: Extract the pre-set action description of each reference feature class for pairwise similarity comparison to obtain the description similarity; If there is no description similarity exceeding a set similarity threshold, the reference feature class with the highest feature coincidence degree is regarded as the actual feature class; If the description similarity between the two reference feature classes exceeds the set similarity threshold, and the description similarity is the maximum value among all similarity comparison results, the current two reference feature classes are respectively labeled as the first feature class and the second feature class; The first feature class is determined as a feature with feature overlap through feature item mapping, and is marked as a first analysis feature; The second feature class is determined as a feature with feature overlap through feature item mapping, and is marked as a second analysis feature; According to the first analysis feature or the second analysis feature, the historical appearance frequency in a preset time period, the video usage rate and the historical video rendering level, the overlap importance score of the first feature class or the second feature class is determined respectively; The corresponding overlap importance score of the first feature class or the second feature class is combined with the feature importance weight to obtain the corresponding comprehensive importance score; The feature class with the maximum comprehensive importance score in the first feature class and the second feature class is taken as the actual feature class and output.

4. The cloud platform-based video production control method of claim 1, wherein, Further comprising: According to the preset division standard, the scene users are divided to obtain user groups, and the user groups are labeled with group categories; Periodically obtain each user group, and the historical video operation data of each action feature class; According to the historical video operation data, the historical operation category contained in each historical video of the action feature class and the historical execution times of the corresponding historical operation category are determined; By combining the preset attention representation weight of each historical operation category contained in the historical video with the corresponding historical execution times, the reference attention score of the current historical video is obtained; The reference attention scores of each historical video of the action feature class are sequentially constructed into a reference attention curve based on the historical video timestamps; The reference attention curve is analyzed to determine the attention change coefficient of the current action feature class; The attention-adjustment coefficient of the corresponding action feature class is determined by comprehensively considering the user proportion of each user group and the corresponding attention change coefficient; Based on the attention-adjustment coefficient, the feature importance weight of the action feature class is initially adjusted to obtain a pre-importance weight; According to the pre-importance weight from large to small, the action feature classes are sorted to generate a list of pending feature classes; According to the scene action stage of the target scene, the corresponding preset feature class importance list is matched, sorted and compared with the list of pending feature classes, and the feature importance weight of each action feature class is determined according to the sorting comparison result.

5. The cloud platform-based video production control method of claim 4, wherein, According to the sorting comparison result, the feature importance weight of each action feature class is determined, including: If the list of pending feature classes and the preset feature class importance list are consistent, the pre-importance weight of the action feature class is taken as the feature importance weight and output; If the list of pending feature classes and the preset feature class importance list are inconsistent, the action feature classes with inconsistent sorting positions are regarded as adjustment feature classes, and the sorting position difference of the adjustment feature classes is determined; Based on the sorting position difference and the preset adjustment unit, the pre-importance weight of the adjustment feature class is adjusted and taken as the feature importance weight and output.

6. The cloud platform-based video production control method of claim 1, wherein, According to the scene action category of the intermediate video, the corresponding video production template is matched, and resources are dynamically allocated from the cloud platform resource pool according to the action priority level to produce the intermediate video, including: According to the scene action category of the current intermediate video, a video production template is extracted from a preset video production template library as an initial reference template; The production design rules of the initial reference template are compared with the actual design requirements to obtain a design comparison result; According to the design comparison result, if there is a design element that does not conform to the design, the design requirement of the design element that does not conform to the design is extracted from the actual design requirement, and the initial reference template is adjusted to obtain an actual reference template; Based on the resources allocated from the cloud platform resource pool using the action priority level, the intermediate video is produced according to the actual reference template to generate a pending requirement video; The quality detection index is used to detect the quality of the pending requirement video, and when the quality detection is qualified, the pending requirement video is stored as an actual requirement video to a predetermined storage location of the cloud platform. 7.A cloud platform-based video production control system, characterized in that, It comprises: A data processing module is used to collect multi-source scene data in a target scene in real time, and dynamically process the original video stream according to preset edge processing rules to generate an intermediate video with attribute tags; An action analysis module is used to determine the scene action category of the intermediate video based on the analysis of the attribute tags, taking into account the feature description and application preference, and determine the action priority level of the scene action category by deeply analyzing the video operation of the user group; A video production module is used to match the corresponding video production template according to the scene action category of the intermediate video, and dynamically allocate resources from the cloud platform resource pool according to the action priority level to produce the intermediate video; The determination of the scene action category of the intermediate video based on the analysis of the attribute tags, taking into account the feature description and application preference, and the determination of the action priority level of the scene action category by deeply analyzing the video operation of the user group, comprises: The attribute tags attached to the intermediate video are content-analyzed and data-converted using a preset analysis algorithm to obtain pre-analysis information; The pre-analysis information is feature-extracted to obtain key analysis features; The key analysis features are mapped to the action feature classes preset for the target scene to obtain action feature classes with feature item coincidence degrees exceeding a set coincidence threshold, and are regarded as reference feature classes; When there are multiple reference feature classes, the actual feature class is determined by comprehensively considering the description similarity and application preference between the reference feature classes; When there is a single reference feature class, it is output as the actual feature class; The key analysis features are input into an action analysis model adapted to the actual feature class to obtain the scene action category of the corresponding intermediate video; The priority evaluation index matched with the scene action category is used to perform priority analysis based on the analyzed attribute tag data to obtain an action priority score; The action priority score is optimized using the feature importance weight of the actual feature class to which the current scene action category belongs to obtain a priority evaluation coefficient; The action priority level of the intermediate video to which the current scene action category belongs is determined according to the priority evaluation coefficient.

Images