Zero-trust video data access privacy protection method and system

Abstract

The invention relates to the technical field of video security and protection, and discloses a zero-trust video data access privacy protection method and system, and the method comprises the steps: carrying out the semantic perception analysis of a video data access request of an access subject, and obtaining a subject identifier, a request access action and target video data; performing identity verification on the access subject, and performing digital signature packaging on a verification result and a security context attribute of the access subject to obtain an identity token; performing access credibility evaluation on the access subject to obtain a real-time credibility; performing privacy sensitivity analysis on the target video data to obtain a content sensitivity label; performing differential desensitization planning on the target video data to obtain a fine-grained desensitization strategy; performing an access control decision on the target video data, and generating a complete access audit log; performing incremental updating on the behavior baseline and the historical access behavior log; according to the invention, the efficiency of zero-trust video data access privacy protection can be improved.

Description

A zero-trust method and system for protecting privacy in video data access Technical Field

[0001] This invention relates to the field of video security technology, and in particular to a zero-trust video data access privacy protection method and system. Background Technology

[0002] Existing video data access privacy protection solutions often rely on single credential verification for authentication, failing to fully integrate security context information such as the access subject's device environment, network status, and session behavior, resulting in insufficient reliability of authentication. Furthermore, trust assessment of access subjects often employs a static configuration model, lacking dynamic analysis that combines historical access behavior with real-time behavioral deviations, making it difficult to accurately identify potential risks during the access process.

[0003] Regarding video data privacy protection, existing de-identification solutions mostly adopt a uniform processing approach, failing to differentiate based on the sensitivity of video content, the access permissions of the users, and real-time trust levels. This leads to either over-identification affecting data usability or under-identification causing privacy leaks. Furthermore, the linkage between access control decisions, de-identification strategies, and audit logs is insufficient, making it impossible to dynamically optimize protection mechanisms based on access behavior. Therefore, how to achieve accurate and dynamic privacy protection while ensuring the usability of video data has become an urgent problem to be solved. Summary of the Invention

[0004] This invention provides a zero-trust video data access privacy protection method and system to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides a zero-trust video data access privacy protection method, comprising: S1, performing semantic-aware parsing on the video data access request of the accessing subject to obtain the subject identifier, the request access action, and the target video data of the video data access request; S2, authenticating the accessing subject based on the subject identifier, and digitally encapsulating the verification result and the security context attribute of the accessing subject to obtain the identity token of the accessing subject; S3, performing normal behavior analysis on the historical access behavior log of the accessing subject to obtain the behavioral baseline of the accessing subject, and performing [further analysis] on the accessing subject based on the identity token, the behavioral baseline, and the historical access behavior log. S4. Perform privacy sensitivity analysis on the target video data to obtain content sensitivity tags for the target video data. Based on the access request action, the real-time trust level, and the content sensitivity tags, perform differentiated desensitization planning on the target video data to obtain a fine-grained desensitization strategy for the target video data. S5. Based on the fine-grained desensitization strategy, make access control decisions on the target video data and generate a complete access audit log for the access control decisions by combining the decision elements of the access control decisions. S6. Based on the complete access audit log, incrementally update the behavior baseline and the historical access behavior log.

[0006] In a preferred embodiment, the step of semantically parsing the video data access request of the accessing subject to obtain the subject identifier, access request action, and target video data of the video data access request includes: performing protocol parsing on the video data access request of the accessing subject to obtain the original request data of the video data access request; performing syntactic structure decoupling on the original request data to obtain the statement components of the original request data; and performing entity intent association on the statement components based on the syntactic dependency relationships between the statement components to obtain the subject identifier, access request action, and target video data of the video data access request.

[0007] In a preferred embodiment, the step of authenticating the access subject based on the subject identifier and digitally encapsulating the verification result and the access subject's security context attributes to obtain the access subject's identity token includes: performing multi-factor credential verification on the access subject based on the subject identifier to obtain the access subject's multi-dimensional verification factors; performing credential logic authentication on the access subject based on the type and number of the multi-dimensional verification factors to obtain the access subject's comprehensive authentication result; collecting the access subject's device fingerprint, network access point information, and current session behavior characteristics in real time when initiating a request to obtain the access subject's security context attributes; binding the comprehensive authentication result and the security context attributes with an integrity signature to obtain the access subject's signature data block; and standardizing and encoding the signature data block to obtain the access subject's identity token.

[0008] In a preferred embodiment, the step of assessing the access credibility of the access subject based on the identity token, the access subject's behavioral baseline, and historical access behavior logs to obtain the access subject's real-time trust level includes: performing behavioral deviation analysis on the identity token and the access subject's behavioral baseline to obtain the access subject's comprehensive behavioral deviation data; performing time-series risk assessment on the comprehensive behavioral deviation data based on the access subject's historical access behavior logs to obtain the access subject's real-time session risk score; and performing trust quantification derivation on the real-time session risk score to obtain the access subject's real-time trust level.

[0009] In a preferred embodiment, the calculation formula for the trust quantification derivation is as follows: In the formula, This indicates the real-time trust level. This indicates the initial baseline trust level of the accessing entity. This represents the real-time session risk score. This represents the arithmetic mean of the historical session risk scores in the historical access behavior logs. This represents the standard deviation of the historical session risk score. This represents the 75th percentile of the historical session risk score. The cumulative distribution function represents the standard normal distribution. Indicators representing the stability of the stated behavioral baseline This represents the natural exponential function. This represents the maximum value function.

[0010] In a preferred embodiment, the step of performing privacy-sensitive analysis on the target video data to obtain content sensitivity tags for the target video data includes: parsing the scene content of the target video data and identifying privacy-sensitive entities in the target video data; performing global tracking and perception of the privacy-sensitive entities based on the spatiotemporal information of the target video data to obtain the occurrence frequency and spatial distribution density of the privacy-sensitive entities; performing sensitivity fusion evaluation on the privacy-sensitive entities based on the occurrence frequency and the spatial distribution density to obtain a global sensitivity score for the target video data; and aggregating the entity type, occurrence frequency, spatial distribution density, and global sensitivity score of the privacy-sensitive entities into content sensitivity tags for the target video data.

[0011] In a preferred embodiment, the step of performing differentiated desensitization planning on the target video data based on the access request action, the real-time trust level, and the content sensitivity tag to obtain a fine-grained desensitization strategy for the target video data includes: mapping the access request action to a preset operation permission rule base to obtain the basic operation permission level of the access request action; determining the privacy level of the content sensitivity tag to obtain the regional sensitivity level of the content sensitivity tag; performing multi-parameter fusion analysis on the regional sensitivity level, the basic operation permission level, and the real-time trust level to obtain the target regional desensitization strength of the target video data; and reconstructing the desensitization strategy for the target video data based on the target regional desensitization strength to obtain a fine-grained desensitization strategy for the target video data.

[0012] In a preferred embodiment, the step of making access control decisions on the target video data based on the fine-grained de-identification strategy, and generating a complete access audit log for the access control decisions by combining the decision elements of the access control decisions, includes: making access control decisions on the target video data according to the fine-grained de-identification strategy to obtain a de-identified video data stream of the target video data; integrating the subject identifier, the access request action, the target video data, the fine-grained de-identification strategy, the real-time trust level, the content sensitivity tag, and the de-identified video data stream into decision elements for the access control decisions; and structuring the decision elements according to a predefined audit data template to obtain a complete access audit log for the access control decisions.

[0013] In a preferred embodiment, the step of performing access control decisions on the target video data according to the fine-grained desensitization strategy to obtain a desensitized video data stream of the target video data includes: extracting directional parameters from the fine-grained desensitization strategy to obtain desensitization control parameters for the target video data; performing frame-level region segmentation on the original video stream of the target video data based on the spatiotemporal coordinate range of the desensitization control parameters to obtain sensitive region image blocks of the original video stream; performing visual information transformation processing on the sensitive region image blocks based on the desensitization operation type and operation intensity parameters of the desensitization control parameters to obtain desensitized image blocks of the sensitive region image blocks; and merging the desensitized image blocks into the original video frame positions of the sensitive region image blocks according to the temporal order of the original video stream to obtain a desensitized video data stream of the target video data.

[0014] To address the aforementioned issues, this invention also provides a zero-trust video data access privacy protection system. The system includes: a request parsing module, used to perform semantic-aware parsing on the video data access request of the accessing subject to obtain the subject identifier, access request action, and target video data of the video data access request; an authentication and token module, used to authenticate the accessing subject based on the subject identifier, and digitally encapsulate the authentication result and the accessing subject's security context attributes to obtain the accessing subject's identity token; and a dynamic trust assessment module, used to assess the access trustworthiness of the accessing subject based on the identity token, the accessing subject's behavioral baseline, and historical access behavior logs to obtain the accessing subject's real-time trust value. The system includes a trust level module; a content awareness and policy generation module, used to perform privacy-sensitive analysis on the target video data, obtain content sensitivity tags for the target video data, and perform differentiated desensitization planning on the target video data based on the requested access action, the real-time trust level, and the content sensitivity tags, to obtain a fine-grained desensitization policy for the target video data; a policy execution and auditing module, used to make access control decisions on the target video data based on the fine-grained desensitization policy, and generate a complete access audit log for the access control decision by combining the decision-making elements of the access control decision; and an adaptive update module, used to incrementally update the behavior baseline and the historical access behavior log based on the complete access audit log.

[0015] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention accurately extracts the subject identifier, access action and target data in video data access requests through semantic perception parsing, combines multi-factor credential verification and security context attributes for digital signature encapsulation, constructs a highly reliable identity token, and at the same time uses a dynamic trust assessment model to quantify real-time trust level, thereby achieving accurate perception and control of access risks, significantly improving the accuracy of identity authentication and risk identification, and strengthening the security of access control.

[0016] 2. This invention generates content sensitivity tags containing multi-dimensional information through privacy sensitivity analysis, and formulates fine-grained desensitization strategies by combining multi-parameter fusion analysis, thereby achieving differentiated privacy protection for video data and balancing privacy security and data availability. By driving the behavior baseline and incremental updates of historical logs through complete access audit logs, the protection mechanism is continuously adaptively optimized, effectively improving the efficiency and flexibility of privacy protection for video data access. Attached Figure Description

[0017] Figure 1 is a flowchart illustrating a zero-trust video data access privacy protection method according to an embodiment of the present invention; Figure 2 is a functional block diagram illustrating a zero-trust video data access privacy protection system according to an embodiment of the present invention; the realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0019] This application provides a zero-trust video data access privacy protection method. The executing entity of this zero-trust video data access privacy protection method includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the method provided in this application embodiment: a server, a terminal, etc. In other words, the zero-trust video data access privacy protection method can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0020] Referring to Figure 1, a flowchart illustrating a zero-trust video data access privacy protection method according to an embodiment of the present invention is shown. In this embodiment, the zero-trust video data access privacy protection method includes: S1, performing semantic-aware parsing on the video data access request of the accessing subject to obtain the subject identifier, access request action, and target video data of the video data access request; in this embodiment, it includes: performing protocol parsing on the video data access request of the accessing subject to obtain the original request data of the video data access request; performing syntactic decoupling on the original request data to obtain the statement components of the original request data; and performing entity intent association on the statement components based on the syntactic dependency relationships between the statement components to obtain the subject identifier, access request action, and target video data of the video data access request.

[0021] For video data access requests initiated by a user, the relevant video data access protocol must first be identified. This protocol includes predefined field format specifications and parsing standards. The field format specifies the order, data length, and encoding format of core fields such as the user identifier field, operation type field, and target data address field. The parsing standards specify the reading method and data verification requirements for each field. Following these protocol specifications, starting from the frame header of the request data, the data content of each protocol frame is read frame by frame. The core content explicitly required in the protocol frame structure, including the access initiation-related identifier segment, operation description segment, and target data pointer segment, is extracted sequentially. During the extraction process, the list of required fields marked in the protocol is strictly checked to ensure that no required field is omitted. Only data within the protocol definition is extracted, without adding any redundant data unrelated to the request. Finally, the original request data for the video data access request, which retains the original state of each field and fully conforms to the protocol specifications, is obtained.

[0022] The pre-defined standard for classifying grammatical components is based on general Chinese grammar rules and specifically formulated in conjunction with the textual expression characteristics of video data access requests. This standard clarifies the criteria and boundaries for determining various grammatical components. Among them, nominal components must be able to refer to specific things or objects and can function as the subject or object in a sentence; verbal components must be able to express specific actions or operations and can function as the predicate in a sentence; adjective components must be able to modify or limit the characteristics of things and are often used in conjunction with nominal components; and relational components must be able to connect different sentence components and clearly reflect the logical relationship between components. According to this standard, the text content of the original request data is first divided into independent sentences according to punctuation marks such as commas and semicolons. Then, the grammatical function of each sentence in the whole sentence is analyzed word by word. Combined with the collocation relationship and semantic association of adjacent words, the grammatical component to which each word belongs is determined. For words with polysemy, the unique grammatical attribute is determined by the specific context of the request text and the logical relationship with other words. This ensures that the grammatical attribute of each sentence component is clear and unique, and there is no overlap of components or ambiguity of attributes. Finally, the independent sentence components after the original request data is split are obtained.

[0023] The pre-established rules for identifying grammatical dependency relationships detail the specific criteria for determining various types of grammatical dependency relationships, such as subject-predicate, verb-object, modifier-head, and coordinate relationships. Specifically, the criteria for a subject-predicate relationship are that the preceding sentence element is a noun and the following sentence element is a verb, and the latter accurately describes the action or state of the former; the criteria for a verb-object relationship are that the preceding sentence element is a verb and the following sentence element is a noun, and the former directly affects the latter; the criteria for a modifier-head relationship are that the preceding sentence element is an adjective and the following sentence element is either a noun or a verb, and the former modifies or limits the characteristics or action of the latter; and the criteria for a coordinate relationship are that two or more sentence elements have the same grammatical attributes and are equal in status within the sentence, with no distinction between primary and secondary elements. Based on this rule, the logical relationships between each sentence component are analyzed one by one. For noun phrases that meet the subject-predicate relationship criteria and contain user account, device number, or other characteristic information clearly indicating the initiator of the access, the noun phrase is identified as the subject identifier. For verb phrases that meet the verb-object relationship criteria and contain information such as download, view, or modify that clearly indicates the specific operation related to access, the verb phrase is identified as the request access action. For noun phrases that meet the verb-object relationship criteria and contain information such as video file name, storage path, or data number that clearly indicates the accessed object, the noun phrase is identified as the target video data. During the association process, the grammatical logic and semantic matching degree between components are repeatedly checked to ensure that each core piece of information corresponds precisely and without deviation to its corresponding sentence component. Ultimately, the subject identifier, request access action, and target video data of the video data access request are obtained.

[0024] The beneficial effects include: a clearly defined protocol parsing standard standardizes the data extraction process, ensuring the integrity and standardization of the original request data and avoiding data loss or redundancy caused by non-standard parsing; fixed syntactic component segmentation rules guarantee the accuracy and consistency of sentence component decomposition, laying a solid foundation for subsequent correlation analysis; and clear syntactic dependency relationship identification logic enables precise correspondence between core information and sentence components, effectively avoiding errors in core information extraction. This series of operations ensures that the parsing process of video data access requests has clear execution standards and reproducibility, ensuring that the extracted subject identifier, request access action, and target video data are complete, accurate, and unambiguous. This provides high-quality basic data support for subsequent key privacy protection processes such as identity verification, dynamic trust assessment, and desensitization strategy formulation, ensuring the stable execution and protective effect of the entire zero-trust video data access privacy protection solution from the source.

[0025] S2. Based on the subject identifier, authenticate the access subject, and encapsulate the verification result and the security context attribute of the access subject with a digital signature to obtain the access subject's identity token; in this embodiment of the invention, it includes: performing multi-factor credential verification on the access subject based on the subject identifier to obtain the access subject's multi-dimensional verification factors; performing credential logic authentication on the access subject based on the type and number of the multi-dimensional verification factors to obtain the access subject's comprehensive authentication result; collecting the access subject's device fingerprint, network access point information, and current session behavior characteristics in real time when initiating a request to obtain the access subject's security context attribute; binding the comprehensive authentication result and the security context attribute with an integrity signature to obtain the access subject's signature data block; and standardizing and encoding the signature data block to obtain the access subject's identity token.

[0026] Based on the acquired entity identifier, a pre-defined multi-factor authentication process is initiated. This process includes three fixed authentication factors: password verification, biometric verification, and hardware token verification. The accessing entity is required to submit the corresponding credentials sequentially. Password verification determines whether the string entered by the accessing entity is completely identical to the encrypted string pre-stored in the system. Biometric verification performs key point matching by extracting fingerprint or facial feature data provided by the accessing entity and matching it with the system's registered feature data. Hardware token verification determines whether the dynamic code entered by the accessing entity is identical to the dynamic code synchronously generated by the system. The original data and verification status records corresponding to each type of credential that passes verification are integrated to obtain the accessing entity's multi-dimensional verification factors.

[0027] The system pre-defines authentication rules for credentials, specifying the effective weight and minimum pass count for each verification factor. For example, password verification has a weight of 30%, biometric verification has a weight of 40%, and hardware token verification has a weight of 30%. At least two verification factors must pass, and the total weight score must be no less than 60 points. Based on these rules, the system first checks the pass status of each type of multi-dimensional verification factor, then calculates the total weight score. If the total score reaches or exceeds 60 points, it is considered "authentication passed" and marked "factors valid." If the total score is below 60 points or fewer than two types pass, it is considered "authentication failed" and marked "insufficient factors" or "weight score not met," ultimately forming the comprehensive authentication result for the access subject.

[0028] The system employs a built-in device information collection tool to extract hardware parameters of the device used by the access subject when initiating a request in real time, including CPU model, memory capacity, hard disk serial number, operating system version, and device unique identifier, and integrates these to generate a device fingerprint. A network protocol analysis tool is used to obtain the access subject's IP address, gateway address, subnet mask, and network access type, forming network access point information. A session monitoring tool records the initiation time, request data size, operation command sequence, and session duration of the current session, extracting current session behavior characteristics. The device fingerprint, network access point information, and current session behavior characteristics are then integrated according to a preset data structure to obtain the access subject's security context attributes.

[0029] The private key in the asymmetric encryption algorithm is used to bind the integrity signature of the comprehensive authentication result and the security context attribute. First, the text information of the comprehensive authentication result and the structured data of the security context attribute are concatenated into a complete data string in the order of "authentication result + device fingerprint + network access point information + session behavior characteristics". Then, the data string is encrypted using a preset private key to generate a fixed-length digital signature. The original concatenated data string and the digital signature are combined in the order of their concatenation to form an immutable signature data block of the access subject.

[0030] The signature data block is standardized and encoded according to the Base64 encoding standard. During the encoding process, the binary data of the signature data block is first grouped into groups of 6 bits each. Each group corresponds to a character in the Base64 encoding table. Any part less than 6 bits is padded with zeros. Then, the corresponding ASCII strings are generated sequentially. The encoded strings must be multiples of 4 in length and do not contain any special characters. Finally, a unified identity token of the access subject that can be directly transmitted and stored is obtained.

[0031] The beneficial effects are that by using multi-dimensional, clearly defined criteria for credential verification and logical authentication, the accuracy and reliability of the comprehensive authentication results are ensured. The system fully collects three types of security context information: device, network, and session, making identity verification more relevant to the access scenario. It adopts asymmetric encryption signature binding and standardized encoding to ensure data integrity and token format uniformity. The generated identity token is both secure and universal, providing a comprehensive and reliable identity basis for subsequent access credibility assessment and strengthening the identity control foundation of the entire privacy protection scheme.

[0032] S3. Perform routine behavior analysis on the historical access behavior logs of the accessing subject to obtain the behavioral baseline of the accessing subject, and based on the identity token, the behavioral baseline, and the historical access behavior logs, evaluate the access credibility of the accessing subject to obtain the real-time trust level of the accessing subject; In this embodiment of the invention, it includes: performing routine behavior analysis on the historical access behavior logs of the accessing subject to obtain the behavioral baseline of the accessing subject; performing behavioral deviation analysis on the identity token and the behavioral baseline of the accessing subject to obtain the comprehensive behavioral deviation data of the accessing subject; performing time-series risk assessment on the comprehensive behavioral deviation data based on the historical access behavior logs of the accessing subject to obtain the real-time session risk score of the accessing subject; and performing trust quantification derivation on the real-time session risk score to obtain the real-time trust level of the accessing subject.

[0033] The calculation formula for the trust quantification derivation is as follows: In the formula, This indicates the real-time trust level. This indicates the initial baseline trust level of the accessing entity. This represents the real-time session risk score. This represents the arithmetic mean of the historical session risk scores in the historical access behavior logs. This represents the standard deviation of the historical session risk score. This represents the 75th percentile of the historical session risk score. The cumulative distribution function represents the standard normal distribution. Indicators representing the stability of the stated behavioral baseline This represents the natural exponential function. This represents the maximum value function.

[0034] Retrieve the historical access behavior logs of the accessing entity over the past 6 months. Extract key behavioral data from the structured fields of the logs, specifically including the timestamp of each access, the device fingerprint used, network access point information, access request actions, and session duration. The extraction process strictly corresponds to the log fields to ensure data completeness and no omissions. Divide the day into 24-hour periods using a 24-hour format, count the number of accesses in each period, calculate the proportion of accesses in each period to the total number of accesses, and identify the continuous period with a cumulative proportion of 70% or higher. This continuous period is the accessing entity's regular access period. Count the number of accesses corresponding to each device fingerprint in the historical access behavior logs, calculate the proportion of accesses for each device to the total number of accesses, and determine the device type corresponding to the device fingerprint with a proportion ≥60% as the accessing entity's frequently used device type. Count the number of accesses from the city of the IP address in the network access point information, calculate the proportion of accesses from each city to the total number of accesses, and determine the city with a proportion ≥50% as the accessing entity's habitual network access area, and record the access type corresponding to this area as the normal access type. Calculate the total number of visits by the accessing entity over the past 6 months, divide by 24 to obtain the average number of visits per week, which is the accessing entity's regular operation frequency. Collect the duration data of all sessions in the historical access behavior logs, arrange them in ascending order, and take the middle value as the baseline for normal session duration. Integrate the above-derived statistics on normal access periods, common device types, habitual network access areas and types, regular operation frequency, and normal session duration baseline to form a behavioral baseline for the accessing entity containing clear characteristics and specific ranges or values. Each characteristic corresponds to the normal performance of historical access behavior, ensuring that the behavioral baseline has clear quantitative basis and reproducibility.

[0035] The system extracts access subject behavior information from the identity token. This information includes behavioral data corresponding to security context attributes such as device fingerprints, network access point information, and session behavior characteristics. The system defines the access subject's behavioral baseline as a feature set established based on the subject's stable access behavior over the past six months, including specific features such as fixed access times, commonly used device types, regular operation frequency, and habitual network access areas. The system compares the behavioral data in the identity token with the corresponding features of the behavioral baseline, setting deviation criteria for each feature. For example: an access time deviation exceeding 3 hours is considered a time-based deviation; any mismatch in CPU model, hard drive serial number, or operating system version in the device fingerprint is considered a device deviation; a network access area not belonging to the same city as the baseline's recorded common area is considered a network deviation; and an operation frequency differing from the baseline's average frequency by more than 50% is considered a frequency deviation. Alternatively, a fixed weight can be assigned to each deviation, such as 2 points for time-based deviation, 3 points for device deviation, 2 points for network deviation, and 1 point for frequency deviation. The sum of the weights corresponding to all deviation features is the overall behavioral deviation data of the access subject.

[0036] The system retrieves the historical access behavior logs of the accessing subject. These logs detail the comprehensive deviation data of each access over the past 12 months, the access results, and whether any security incidents were triggered. A risk scoring mapping rule is established based on these logs. For example, a comprehensive deviation data score of 0-2 corresponds to a basic risk score of 10, 3-5 corresponds to 30, 6-8 corresponds to 60, and 9 and above corresponds to 90. The number of security incidents triggered within each basic risk score range in the historical logs is counted, and the probability of security incidents occurring in each range is calculated. For example, if the probability of a security incident in the 0-2 range is below 5%, the basic score is maintained; if the probability in the 3-5 range is between 5% and 15%, the basic score is increased by 10 points; if the probability in the 6-8 range is between 15% and 30%, the basic score is increased by 20; and if the probability in the 9 and above range is above 30%, the basic score is increased by 30. Based on the currently obtained comprehensive deviation data, the corresponding basic risk score range is determined. The basic score is then adjusted according to the probability of security incidents within that range. The final value obtained is the real-time session risk score of the accessing subject.

[0037] Determine the initial baseline trust level for the accessing entity. For example, the initial baseline trust level for accessing entities with complete registration information and passed qualification verification is uniformly set to 90 points. Collect all historical session risk scores from historical access behavior logs, add these scores together, and divide by the total number of historical session risk scores to obtain the arithmetic mean of the historical session risk scores. Calculate the difference between each historical session risk score and the arithmetic mean, square each difference, sum them up, divide the sum by the total number of historical session risk scores to obtain the variance, and take the square root of this variance to obtain the standard deviation of the historical session risk scores.

[0038] All historical session risk scores are arranged in ascending order. The total number of historical session risk scores is counted and multiplied by 75% to obtain the location index. If the location index is an integer, the score corresponding to that index is taken as the 75th percentile of the historical session risk score; if the location index is not an integer, it is rounded up and the score corresponding to that index is taken as the 75th percentile. The difference between the real-time session risk score and the arithmetic mean of the historical session risk scores is calculated. This difference is divided by the standard deviation of the historical session risk scores to obtain a ratio. Based on the position of this ratio on the standard normal distribution curve, the cumulative distribution function result of the standard normal distribution is determined. This result is between 0 and 1. For example, the number of adjustments to the baseline behavior of the visiting subject in the past 3 months can be counted. When the number of adjustments is 0, the stability index of the behavioral baseline is set to 1.0; when the number of adjustments is 1, it is set to 0.8; when the number of adjustments is 2, it is set to 0.6; and when the number of adjustments exceeds 2, it is set to 0.4. Calculate the difference between the real-time session risk score and the 75th percentile of the historical session risk score. If the difference is positive, the maximum function result is the difference; if the difference is negative or 0, the maximum function result is 0. Subtract the stability index of the behavioral baseline from 1, multiply the result by the maximum function result, divide the product by 2, take the negative of the result as the exponent, and perform a power operation with the natural constant e as the base to obtain the natural exponential function result. Multiply the initial baseline trust score by 1 and subtract the cumulative distribution function result of the standard normal distribution, then multiply the product by the natural exponential function result. The final value is the real-time trust score of the access subject.For example, if the accessing entity's registration information is complete and has passed the qualification review, the initial baseline trust level is set at 90 points. Retrieving its historical access behavior logs from the past 12 months, eight historical session risk scores are extracted: 10, 20, 30, 40, 50, 60, 70, and 80 points. These eight scores are summed to obtain 360 points, then divided by 8 to obtain the arithmetic mean of 45 points. The difference between each score and 45 points is calculated, squared, and summed to obtain 4200. Dividing this by 8 gives the variance of 525. Taking the square root of 525 gives the standard deviation of approximately 22.91 points. Multiplying the total number of data points by 75% yields the position index 6. Arranging these in ascending order, the 6th data point is selected, determining the 75th percentile as 60 points. Based on the current behavior, a comprehensive deviation from the data and a time-series risk assessment are conducted. The real-time session risk score is 55. Subtracting 45 from 55 (a difference of 10 points) and dividing by 22.91 yields a ratio of approximately 0.436. Based on the corresponding position of this ratio on the standard normal distribution curve, the cumulative distribution function result is determined to be 0.669. The number of adjustments to the baseline behavior of this access subject in the past 3 months is 0, corresponding to a stability index set to 1.0. The difference between 55 and 60 is -5 points, so the maximum value function result is 0. Subtracting 1.0 from 1 yields 0, multiplying by 0 and dividing by 2 still results in 0. The negative number of this result is taken as the exponent, and exponentiation is performed with the natural constant e as the base, resulting in a natural exponent function result of 1. Finally, multiplying 90 by 1 and subtracting 0.669, then multiplying by 1, yields a real-time trust score of 29.80 for the access subject.

[0039] The beneficial effects include: achieving precise quantification of behavioral deviations through clear feature comparison standards and weight settings; the resulting comprehensive behavioral deviation data objectively reflecting the degree of difference between the access behavior and the baseline; risk scoring rules and probability adjustment mechanisms based on historical logs ensuring that real-time session risk scores can accurately match the current access risk level; and a trust quantification derivation process supported by multi-dimensional data, combining historical behavioral patterns, baseline stability, and real-time risk conditions, enabling real-time trust levels to comprehensively and accurately reflect the current trustworthiness of the accessing subject. The entire evaluation process has clear execution standards and reproducible calculation logic, providing a reliable trust basis for subsequent de-identification strategy formulation and access control decisions, effectively enhancing the accuracy and security of video data access privacy protection under the zero-trust architecture.

[0040] S4. Perform privacy-sensitive analysis on the target video data to obtain content sensitivity tags for the target video data, and based on the request access action, the real-time trust level, and the content sensitivity tags, perform differentiated desensitization planning on the target video data to obtain a fine-grained desensitization strategy for the target video data; In this embodiment of the invention, performing privacy-sensitive analysis on the target video data to obtain content sensitivity tags for the target video data includes: performing scene content parsing on the target video data and identifying privacy-sensitive entities in the target video data; performing global tracking and perception on the privacy-sensitive entities based on the spatiotemporal information of the target video data to obtain the occurrence frequency and spatial distribution density of the privacy-sensitive entities; performing sensitivity fusion evaluation on the privacy-sensitive entities based on the occurrence frequency and spatial distribution density to obtain a global sensitivity score for the target video data; and aggregating the entity type, occurrence frequency, spatial distribution density, and global sensitivity score of the privacy-sensitive entities into content sensitivity tags for the target video data.

[0041] The step of performing differentiated desensitization planning on the target video data based on the access request action, the real-time trust level, and the content sensitivity tag to obtain a fine-grained desensitization strategy for the target video data includes: mapping the access request action to a preset operation permission rule base to obtain the basic operation permission level of the access request action; determining the privacy level of the content sensitivity tag to obtain the regional sensitivity level of the content sensitivity tag; performing multi-parameter fusion analysis on the regional sensitivity level, the basic operation permission level, and the real-time trust level to obtain the target regional desensitization strength of the target video data; and reconstructing the desensitization strategy for the target video data based on the target regional desensitization strength to obtain a fine-grained desensitization strategy for the target video data.

[0042] The target video data is analyzed frame by frame, with each frame adhering to a predefined scene recognition standard. This standard specifies image feature thresholds for various scenes, including indoor scenes, outdoor public scenes, and outdoor private scenes. For example, indoor scenes require a light intensity between 100-500 lux and the inclusion of fixed structural features such as walls, doors, and windows; outdoor public scenes require a field of view of ≥60% and the inclusion of public facilities. Simultaneously, entity recognition is performed based on a predefined privacy-sensitive entity feature library. This library stores image and text features of sensitive entities such as faces, license plates, ID card numbers, and bank card numbers. By comparing the image elements and text information in each frame of video data with the features in the library, entities with a feature match of ≥90% are identified as privacy-sensitive entities. Ultimately, all privacy-sensitive entities in the target video data are identified.

[0043] The spatiotemporal information of the target video data is extracted. The temporal information is the timestamp corresponding to each frame of video, and the spatial information is the coordinate range of pixels within each frame. Based on the timestamp, time intervals are divided by minutes. The frequency of occurrence of each privacy-sensitive entity within each time interval is counted, and this frequency is taken as the occurrence frequency of the privacy-sensitive entity. For example, if a privacy-sensitive entity appears 30 times in 10 minutes, its occurrence frequency is 3 times / minute. Simultaneously, the pixel coverage area of ​​each privacy-sensitive entity across all appearing frames is calculated, and the proportion of this coverage area to the total pixel area of ​​a single frame is calculated as the spatial distribution density of the privacy-sensitive entity. For example, if a privacy-sensitive entity covers an average of 25% of the pixel area of ​​a single frame across 20 appearing frames, its spatial distribution density is 25%.

[0044] The pre-defined sensitivity fusion evaluation rules specify a weight of 0.4 for frequency of occurrence and 0.6 for spatial distribution density. Frequency of occurrence is divided into three levels: ≥5 occurrences / minute (high frequency, 4 points); 2-4 occurrences / minute (medium frequency, 2 points); and ≤1 occurrence / minute (low frequency, 0 points). Spatial distribution density is also divided into three levels: ≥30% (high density, 6 points); 10%-29% (medium density, 3 points); and ≤9% (low density, 0 points). Based on the scores corresponding to these levels, a weighted sum is calculated using the pre-defined weights: Global Sensitivity Score = Frequency of Occurrence Score × 0.4 + Spatial Distribution Density Score × 0.6, resulting in the final global sensitivity score for the target video data. For example, if a license plate is identified as a privacy-sensitive entity in the target video data, after extracting the spatiotemporal information of the target video data, the frequency of the license plate occurrence is calculated to be 4 times / minute, which is a medium frequency level corresponding to 2 points. At the same time, the pixel coverage of the license plate in all occurrence frames is calculated to account for 28% of the total pixel area of ​​a single frame, which is a medium density level corresponding to 3 points. According to the preset sensitivity fusion evaluation rules, the occurrence frequency weight is 0.4 and the spatial distribution density weight is 0.6. By calculating 2 points × 0.4 + 3 points × 0.6 = 0.8 points + 1.8 points = 2.6 points, the final global sensitivity score of the target video data is 2.6 points.

[0045] The specific types of privacy-sensitive entities identified, their statistical frequency of occurrence, their calculated spatial distribution density, and their weighted global sensitivity scores are integrated according to a fixed format of "entity type - frequency of occurrence - spatial distribution density - global sensitivity score" to form a content sensitivity label for the target video data with a unified structure and complete information.

[0046] A pre-defined operation permission rule base is established, clearly defining four basic operation permission levels: Level 1 corresponds to viewing, Level 2 to downloading, Level 3 to modifying, and Level 4 to uploading. Each permission level has a clearly defined scope of operations. The system precisely matches the obtained access request with the operation type in the rule base. If the request is to view, it matches Level 1 permission; if it's to download, it matches Level 2 permission; if it's to modify, it matches Level 3 permission; and if it's to upload, it matches Level 4 permission, ultimately determining the basic operation permission level corresponding to the requested access action.

[0047] A preset privacy level assessment standard is established: a global sensitivity score between 0 and 3 is classified as low sensitivity; a score between 4 and 6 is classified as medium sensitivity; and a score between 7 and 10 is classified as high sensitivity. Based on this standard, the global sensitivity scores of the acquired content sensitivity tags are matched to determine the privacy level corresponding to the target video data, ultimately yielding the regional sensitivity level of the content sensitivity tags.

[0048] A multi-parameter fusion analysis rule is preset, setting the weight of regional sensitivity level to 0.5, basic operation permission level to 0.3, and real-time trust level to 0.2. Regional sensitivity levels are converted into corresponding scores: low sensitivity level corresponds to 1 point, medium sensitivity level to 2 points, and high sensitivity level to 3 points. Basic operation permission levels are also converted into corresponding scores: Level 1 permission corresponds to 1 point, Level 2 permission to 2 points, Level 3 permission to 3 points, and Level 4 permission to 4 points. Real-time trust level is converted into corresponding scores within the range of 0-100: 0-25 points correspond to 1 point, 26-50 points to 2 points, 51-75 points to 3 points, and 76-100 points to 4 points. The weighted sum of the converted scores for the three parameters is calculated as follows: Fusion Score = Regional Sensitivity Level Score × 0.5 + Basic Operation Permission Level Score × 0.3 + Real-Time Trust Level Score × 0.2. When the fusion score is between 1 and 2, the desensitization intensity of the target area is determined to be weak; when the fusion score is between 3 and 5, it is determined to be medium desensitization; when the fusion score is between 6 and 9, it is determined to be strong desensitization, and the desensitization intensity of the target area of ​​the target video data is finally obtained.

[0049] A pre-defined desensitization strategy library contains specific operation methods and parameters corresponding to three levels of desensitization intensity: weak, medium, and strong. Weak desensitization intensity corresponds to pixelation processing, with a pixelation level set to 20%, meaning 20% ​​of the pixels in the sensitive area are blurred. Medium desensitization intensity corresponds to partial occlusion processing, where the occluded area occupies 80% of the total area of ​​the sensitive entity, using color blocks consistent with the surrounding scene color tone. Strong desensitization intensity corresponds to complete replacement processing, completely covering the sensitive area with a pure white color block. Based on the determined desensitization intensity of the target area, the corresponding desensitization operation method and parameters are retrieved from the desensitization strategy library. Combined with the location information of privacy-sensitive entities in the target video data, a separate desensitization scheme is formulated for each sensitive area, ultimately resulting in a fine-grained desensitization strategy for the target video data.

[0050] The beneficial effects include: accurate identification of privacy-sensitive entities through clear scene analysis standards and a sensitive entity feature library; reproducible calculation of frequency and spatial distribution density based on spatiotemporal information quantitative statistics; a weighted fusion evaluation mechanism to ensure that the global sensitivity score objectively reflects the degree of video privacy sensitivity; differentiated desensitization intensity based on multi-parameter fusion analysis combined with access scenario and subject trust level; and fine-grained desensitization strategies to accurately match desensitization methods for different sensitive areas, effectively preventing privacy leakage risks while maximizing the usability of video data, providing a scientific and reliable basis for subsequent access control decisions, and improving the accuracy and flexibility of privacy protection for zero-trust video data access.

[0051] S5. Based on the fine-grained desensitization strategy, perform access control decisions on the target video data, and generate a complete access audit log for the access control decisions by combining the decision elements of the access control decisions; in this embodiment of the invention, the following steps are included: performing access control decisions on the target video data according to the fine-grained desensitization strategy to obtain a desensitized video data stream of the target video data; integrating the subject identifier, the access request action, the target video data, the fine-grained desensitization strategy, the real-time trust level, the content sensitivity tag, and the desensitized video data stream into decision elements for the access control decisions; and structuring the decision elements according to a predefined audit data template to obtain a complete access audit log for the access control decisions.

[0052] The step of performing access control decisions on the target video data according to the fine-grained desensitization strategy to obtain the desensitized video data stream of the target video data includes: extracting directional parameters from the fine-grained desensitization strategy to obtain desensitization control parameters for the target video data; performing frame-level region segmentation on the original video stream of the target video data based on the spatiotemporal coordinate range of the desensitization control parameters to obtain sensitive region image blocks of the original video stream; performing visual information transformation processing on the sensitive region image blocks based on the desensitization operation type and operation intensity parameters of the desensitization control parameters to obtain desensitized image blocks of the sensitive region image blocks; and fusing the desensitized image blocks into the original video frame positions of the sensitive region image blocks according to the temporal order of the original video stream to obtain the desensitized video data stream of the target video data.

[0053] The fine-grained desensitization strategy is subjected to targeted parameter extraction. This strategy explicitly includes three core information categories: the spatiotemporal coordinate range corresponding to the desensitization operation, the desensitization operation type, and the operation intensity parameter. During the extraction process, the specific values ​​and descriptions of each type of information are extracted one by one. The spatiotemporal coordinate range needs to be accurate to the video frame number and the pixel coordinate range within each frame. The desensitization operation type is clearly defined as pixelation, partial occlusion, or complete replacement. The operation intensity parameter corresponds to the specific processing degree standard. Finally, the desensitization control parameters of the target video data are obtained.

[0054] Based on the spatiotemporal coordinate range in the desensitization control parameters, the time coordinates in the spatiotemporal coordinate range determine the frame number interval of the video to be processed, and the spatial coordinates determine the upper left and lower right corner coordinates of the pixels of the sensitive area in each frame. The original video stream of the target video data is processed frame by frame according to the frame number order. The area in each frame of the video that conforms to the spatial coordinate interval is precisely cut. The cutting process strictly follows the pixel coordinate boundary and does not exceed or reduce the specified range. Finally, sensitive area image blocks corresponding to all sensitive areas in the original video stream are obtained.

[0055] Based on the desensitization operation type and intensity parameters in the desensitization control parameters, if the desensitization operation type is pixelation, the pixel block size is set according to the intensity parameter: intensity 1 corresponds to a 3×3 pixel block, intensity 2 corresponds to a 5×5 pixel block, and intensity 3 corresponds to a 7×7 pixel block. The sensitive area image block is then reassembled according to the set pixel block size to achieve visual blurring. If it is partial occlusion, the occlusion area ratio is determined according to the intensity parameter: intensity 1 is 50%, intensity 2 is 80%, and intensity 3 is 100%. A solid color block with the same color tone as the surrounding pixels of the sensitive area is used for occlusion, and the color tone of the block is determined by extracting the average color tone within a 10-pixel range of the surrounding area. If it is complete replacement, a white solid color block is directly used to cover the entire sensitive area image block. Regardless of the operation type, the standard corresponding to the intensity parameter is strictly followed to obtain the desensitized image block of the sensitive area image block.

[0056] Following the frame sequence of the original video stream, each desensitized image block is placed back into its original position within the video frame where the original sensitive area image block was located. During the placement process, it is ensured that the pixel coordinates of the desensitized image block are completely aligned with the original sensitive area, without changing the content of other non-sensitive areas of the video frame or the time order between frames. After all frames are processed, they are integrated according to the encoding format of the original video stream to finally obtain the desensitized video data stream of the target video data.

[0057] The subject identifier, access request action, original information of target video data, complete content of fine-grained de-identification strategy, specific value of real-time trust level, all information of content sensitivity tag, and storage path and core characteristics of de-identified video data stream obtained in the previous steps are integrated in a fixed order of "subject information - request information - data information - strategy information - trust level information - tag information - de-identification result information" to ensure that each information item is complete and without omission, and finally form the decision elements of access control decision.

[0058] The predefined audit data template contains fixed fields, specifically: access subject identifier, request access action, target video data identifier, fine-grained de-identification strategy, real-time trust level, content sensitivity tag, de-identified video data stream, access control decision time, and decision result. Following the field order of this template, each piece of information from the decision-making element is filled into the corresponding field. The access control decision time field records the current system time, and the decision result field is marked "De-identification completed." During the filling process, ensure accurate matching of information and fields, and maintain a consistent and standardized format to ultimately obtain a complete access audit log for access control decisions.

[0059] The beneficial effects are as follows: precise location and separation of sensitive areas are achieved through explicit parameter extraction and frame-level region segmentation; visual information transformation processing performed according to standards ensures that the desensitization effect meets preset requirements, protecting privacy and security without damaging non-sensitive content; the elements of the decision are integrated in a fixed order and logs are arranged according to standardized templates, ensuring the integrity and uniformity of audit log information, and clearly tracing the entire process of access control decisions; the entire process has clear execution standards and reproducibility, the generated desensitized video data stream accurately meets privacy protection needs, and the complete access audit log provides a reliable basis for subsequent behavior baseline updates and security tracing, further enhancing the security and traceability of zero-trust video data access privacy protection.

[0060] In this embodiment of the invention, step S6 may also be included: incrementally updating the behavior baseline and the historical access behavior log based on the complete access audit log.

[0061] The complete access audit log is parsed to extract key data related to the access subject's behavior and access process. Specifically, this includes the access subject's identifier, access timestamp, device fingerprint used, network access point information, access request action, real-time trust level value, content sensitivity tag, fine-grained de-identification strategy, and de-identified video data stream characteristics. All data is directly extracted from the corresponding structured fields of the audit log to ensure that the extracted information is completely consistent with the log records, without omissions or errors.

[0062] Following the preset storage format of historical access behavior logs, and using "access subject identifier - access timestamp" as a unique index, the extracted key data is structured and organized, ensuring that the organized data fields completely match the fields in the historical access behavior logs. First, the unique index is used to check if a duplicate record exists in the historical access behavior logs. If not, a new complete access record is added directly; if it exists, dynamically changing data in the original record, such as real-time trust level and anonymization strategy execution results, is overwritten, while static basic data, such as the subject identifier, is retained. This ensures that the historical access behavior logs are updated with the latest access information in real time while maintaining data uniqueness.

[0063] The update difference thresholds for each feature item of the preset behavior baseline are set. For example, the difference thresholds for four types of feature items, namely access time period, common device type, regular operation frequency and habitual network access area, are all set to 10%. That is, when the statistical result after the integration of new access data differs from the original feature item by more than 10%, the feature item update is initiated.

[0064] After statistically integrating newly added access records, the behavioral data of the accessing entity over the past 6 months is used to calculate the distribution ratio of access time periods, the usage frequency ratio of each device, the average operation frequency, and the access frequency ratio of each network access area. These statistical results are then compared one by one with the corresponding features of the original behavioral baseline. If the statistical result of a certain feature differs from the original baseline by more than 10%, the original feature is replaced with the new statistical result; if the difference does not exceed 10%, the original feature remains unchanged, completing the incremental update of the behavioral baseline.

[0065] The beneficial effects include ensuring the reliability of updated data sources through precise analysis of audit logs; maintaining the integrity and uniqueness of historical access behavior logs through a unique indexing mechanism to avoid data redundancy or duplication; and a behavior baseline update mechanism based on preset thresholds to enable the baseline to dynamically adapt to changes in the behavior of the accessing subject. The entire incremental update process is standardized, reproducible, and keeps the behavior baseline and historical access behavior logs up-to-date. This provides realistic basic data for subsequent core processes such as access credibility assessment and de-identification strategy optimization, continuously improving the adaptability and protection accuracy of the zero-trust video data access privacy protection system.

[0066] Figure 2 shows a functional block diagram of a zero-trust video data access privacy protection system provided in an embodiment of the present invention.

[0067] The zero-trust video data access privacy protection system 100 described in this invention can be installed in an electronic device. Depending on the functions implemented, the zero-trust video data access privacy protection system 100 may include a request parsing module 101, an authentication and token module 102, a dynamic trust assessment module 103, a content awareness and policy generation module 104, a policy execution and auditing module 105, and an adaptive update module 106. The modules described in this invention can also be referred to as units, which are a series of computer program segments that can be executed by the processor of an electronic device and perform a fixed function, stored in the memory of the electronic device.

[0068] In this embodiment, the functions of each module / unit are as follows: The request parsing module 101 is used to perform semantic-aware parsing on the video data access request of the accessing subject to obtain the subject identifier, the request access action, and the target video data of the video data access request; The authentication and token module 102 is used to authenticate the accessing subject based on the subject identifier, and encapsulate the authentication result and the security context attribute of the accessing subject with a digital signature to obtain the identity token of the accessing subject; The dynamic trust assessment module 103 is used to assess the access credibility of the accessing subject based on the identity token, the behavioral baseline of the accessing subject, and historical access behavior logs to obtain the real-time trust level of the accessing subject; The content awareness... The policy generation module 104 is used to perform privacy-sensitive analysis on the target video data to obtain content sensitivity tags for the target video data, and to perform differentiated desensitization planning on the target video data based on the requested access action, the real-time trust level, and the content sensitivity tags to obtain a fine-grained desensitization strategy for the target video data; the policy execution and auditing module 105 is used to make access control decisions on the target video data based on the fine-grained desensitization strategy, and to generate a complete access audit log for the access control decision by combining the decision-making elements of the access control decision; the adaptive update module 106 is used to incrementally update the behavior baseline and the historical access behavior log based on the complete access audit log.

[0069] In the several embodiments provided by this invention, it should be understood that the disclosed methods and systems can be implemented in other ways. For example, the system embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0070] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0071] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0072] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0073] This application embodiment can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A zero-trust method for protecting privacy in video data access, characterized in that, The method includes: S1, performing semantic-aware parsing on the video data access request of the accessing subject to obtain the subject identifier, access request action, and target video data of the video data access request; S2, authenticating the accessing subject based on the subject identifier, and digitally encapsulating the verification result and the security context attribute of the accessing subject to obtain the identity token of the accessing subject; S3, performing normal behavior analysis on the historical access behavior log of the accessing subject to obtain the behavior baseline of the accessing subject, and evaluating the access credibility of the accessing subject based on the identity token, the behavior baseline, and the historical access behavior log to obtain the accessing subject's identity token. S4. Perform privacy sensitivity analysis on the target video data to obtain content sensitivity tags for the target video data, and based on the request access action, the real-time trust level, and the content sensitivity tags, perform differentiated desensitization planning on the target video data to obtain a fine-grained desensitization strategy for the target video data; S5. Based on the fine-grained desensitization strategy, make access control decisions on the target video data, and generate a complete access audit log for the access control decision by combining the decision-making elements of the access control decision; S6. Based on the complete access audit log, incrementally update the behavior baseline and the historical access behavior log.

2. The zero-trust video data access privacy protection method as described in claim 1, characterized in that, The step of semantically parsing the video data access request of the accessing subject to obtain the subject identifier, access request action, and target video data of the video data access request includes: performing protocol parsing on the video data access request of the accessing subject to obtain the original request data of the video data access request; performing syntactic structure decoupling on the original request data to obtain the statement components of the original request data; and performing entity intent association on the statement components based on the syntactic dependency relationship between the statement components to obtain the subject identifier, access request action, and target video data of the video data access request.

3. The zero-trust video data access privacy protection method as described in claim 1, characterized in that, The process of authenticating the access subject based on the subject identifier and digitally encapsulating the verification result and the access subject's security context attributes to obtain the access subject's identity token includes: performing multi-factor credential verification on the access subject based on the subject identifier to obtain the access subject's multi-dimensional verification factors; performing credential logic authentication on the access subject based on the type and number of the multi-dimensional verification factors to obtain the access subject's comprehensive authentication result; collecting the access subject's device fingerprint, network access point information, and current session behavior characteristics in real time when initiating the request to obtain the access subject's security context attributes; binding the comprehensive authentication result and the security context attributes with an integrity signature to obtain the access subject's signature data block; and standardizing and encoding the signature data block to obtain the access subject's identity token.

4. The zero-trust video data access privacy protection method as described in claim 1, characterized in that, The identity token, the behavioral baseline, and the historical access behavior logs are used to assess the access credibility of the accessing entity and obtain the real-time trust level of the accessing entity. This includes: performing behavioral deviation analysis on the identity token and the behavioral baseline of the accessing entity to obtain comprehensive behavioral deviation data of the accessing entity; performing time-series risk assessment on the comprehensive behavioral deviation data based on the historical access behavior logs of the accessing entity to obtain a real-time session risk score of the accessing entity; and performing trust quantification derivation on the real-time session risk score to obtain the real-time trust level of the accessing entity.

5. The zero-trust video data access privacy protection method as described in claim 4, characterized in that, The calculation formula for the trust quantification derivation is as follows: In the formula, This indicates the real-time trust level. This indicates the initial baseline trust level of the accessing entity. This represents the real-time session risk score. This represents the arithmetic mean of the historical session risk scores in the historical access behavior logs. This represents the standard deviation of the historical session risk score. This represents the 75th percentile of the historical session risk score. The cumulative distribution function represents the standard normal distribution. This represents a stability index of the stated behavioral baseline. This represents the natural exponential function. This represents the maximum value function.

6. The zero-trust video data access privacy protection method as described in claim 1, characterized in that, The step of performing privacy-sensitive analysis on the target video data to obtain content sensitivity tags for the target video data includes: parsing the scene content of the target video data and identifying privacy-sensitive entities in the target video data; performing global tracking and perception of the privacy-sensitive entities based on the spatiotemporal information of the target video data to obtain the occurrence frequency and spatial distribution density of the privacy-sensitive entities; performing sensitivity fusion evaluation on the privacy-sensitive entities based on the occurrence frequency and spatial distribution density to obtain a global sensitivity score for the target video data; and aggregating the entity type, occurrence frequency, spatial distribution density, and global sensitivity score of the privacy-sensitive entities into content sensitivity tags for the target video data.

7. The zero-trust video data access privacy protection method as described in claim 1, characterized in that, The step of performing differentiated desensitization planning on the target video data based on the access request action, the real-time trust level, and the content sensitivity tag to obtain a fine-grained desensitization strategy for the target video data includes: mapping the access request action to a preset operation permission rule base to obtain the basic operation permission level of the access request action; determining the privacy level of the content sensitivity tag to obtain the regional sensitivity level of the content sensitivity tag; performing multi-parameter fusion analysis on the regional sensitivity level, the basic operation permission level, and the real-time trust level to obtain the target regional desensitization strength of the target video data; and reconstructing the desensitization strategy for the target video data based on the target regional desensitization strength to obtain a fine-grained desensitization strategy for the target video data.

8. The zero-trust video data access privacy protection method as described in claim 1, characterized in that, The process of making access control decisions on the target video data based on the fine-grained de-identification strategy, and generating a complete access audit log for the access control decisions by combining the decision elements of the access control decisions, includes: making access control decisions on the target video data according to the fine-grained de-identification strategy to obtain a de-identified video data stream of the target video data; integrating the subject identifier, the access request action, the target video data, the fine-grained de-identification strategy, the real-time trust level, the content sensitivity tag, and the de-identified video data stream into decision elements of the access control decisions; and structuring the decision elements according to a predefined audit data template to obtain a complete access audit log for the access control decisions.

9. A zero-trust video data access privacy protection method as described in claim 8, characterized in that, The step of performing access control decisions on the target video data according to the fine-grained desensitization strategy to obtain the desensitized video data stream of the target video data includes: extracting directional parameters from the fine-grained desensitization strategy to obtain desensitization control parameters for the target video data; performing frame-level region segmentation on the original video stream of the target video data based on the spatiotemporal coordinate range of the desensitization control parameters to obtain sensitive region image blocks of the original video stream; performing visual information transformation processing on the sensitive region image blocks based on the desensitization operation type and operation intensity parameters of the desensitization control parameters to obtain desensitized image blocks of the sensitive region image blocks; and fusing the desensitized image blocks into the original video frame positions of the sensitive region image blocks according to the temporal order of the original video stream to obtain the desensitized video data stream of the target video data.

10. A zero-trust video data access privacy protection system, characterized in that, To implement a zero-trust video data access privacy protection method according to any one of claims 1-9, the system comprises: a request parsing module, configured to perform semantic-aware parsing on the video data access request of the accessing subject to obtain the subject identifier, the request access action, and the target video data of the video data access request; an authentication and token module, configured to authenticate the accessing subject based on the subject identifier, and digitally encapsulate the authentication result and the security context attribute of the accessing subject to obtain the identity token of the accessing subject; and a dynamic trust assessment module, configured to assess the access trustworthiness of the accessing subject based on the identity token, the behavioral baseline of the accessing subject, and historical access behavior logs to obtain the real trust of the accessing subject. The system comprises: a real-time trust level module; a content awareness and policy generation module, used to perform privacy-sensitive analysis on the target video data, obtain content sensitivity tags for the target video data, and perform differentiated desensitization planning on the target video data based on the requested access action, the real-time trust level, and the content sensitivity tags, to obtain a fine-grained desensitization policy for the target video data; a policy execution and auditing module, used to make access control decisions on the target video data based on the fine-grained desensitization policy, and generate a complete access audit log for the access control decision based on the decision-making elements of the access control decision; and an adaptive update module, used to incrementally update the behavior baseline and the historical access behavior log based on the complete access audit log.

Images