Human body posture data access system and method based on hierarchical encryption and dynamic access control

Abstract

The invention belongs to the technical field of computer information security, and particularly relates to a human body posture data access system and method based on hierarchical encryption and dynamic access control. The method is executed by a data management platform server, and specifically comprises the following steps: receiving original human body posture data, and dynamically dividing the original human body posture data into a plurality of data layers with different sensitivity levels based on a multi-dimensional privacy sensitivity quantification result of the data; formulating different access control strategies for each data layer, and executing corresponding encryption operation on each layer of data; and executing identity verification and authorization operation based on the access request. The invention provides a human body posture data privacy protection management scheme which is balanced in control granularity and data utility, balanced in authority management and security demand and compliant in law.

Description

Technical Field

[0001] This application belongs to the field of computer information security technology, specifically relating to a human posture data access system and method based on layered encryption and dynamic access control.

[0002] Systems and methods. Background Technology

[0003] The widespread adoption of remote monitoring technology in nursing homes, postpartum care centers, and other settings has made human posture data a core component of behavioral analysis. However, this type of data, rich in biometric information, faces serious privacy risks during collection, storage, and use, posing a significant challenge to existing security technologies.

[0004] While early methods of directly transmitting raw data ensured data integrity, they were prone to privacy leaks due to a lack of access control. At the same time, the massive amount of data also brought high transmission latency and storage costs.

[0005] To address these shortcomings, the industry has proposed various improvement solutions, but several deep-seated contradictions remain, the core of which lies in the imbalance between privacy protection, data utility, and system performance: Contradiction 1: The contradiction between control granularity and data utility. Existing technical solutions are rather crude in access control, lacking the fine-grained ability to differentiate control based on data sensitivity. For example, relying on data anonymization methods and carrying out "one-size-fits-all" excessive anonymization in pursuit of security can irreversibly damage the original accuracy and utility of the data.

[0006] Contradiction 2: The contradiction between rigid access control and dynamic security requirements. Traditional access control uses a static permission model, where permissions are granted and remain valid indefinitely, unable to be dynamically adjusted based on context such as visitor identity, data sensitivity, and urgency of the operation. This rigid mechanism leads to a dual drawback: Inaccurate permission binding: It is impossible to establish temporary permission associations between highly sensitive data and specific wards (such as emergency access by immediate family members). Lack of emergency response: There is a lack of dynamic permission escalation mechanisms for emergency scenarios.

[0007] The essence of this problem lies in the structural contradiction between the static nature of access control and the dynamic nature of security scenarios, which restricts the system's adaptability and security in sensitive data scenarios.

[0008] Contradiction 3: The contradiction between technological reliability and legal compliance. Existing solutions generally lack end-to-end auditability, fail to record the entire lifecycle of data access in an immutable manner, and cannot provide a complete chain of evidence that complies with regulations such as GDPR, resulting in insufficient system transparency and credibility.

[0009] In summary, existing technologies have multiple contradictions in key dimensions such as privacy protection strength, data availability, and compliance auditability, and cannot meet the comprehensive high requirements for data security, value preservation, and compliance management in sensitive monitoring scenarios. Summary of the Invention

[0010] This application proposes a human posture data privacy protection management system and method based on layered encryption and dynamic access control, aiming to solve the problems of imbalance between control granularity and data utility, imbalance between permission management and security requirements, and imbalance between technical reliability and legal compliance in the prior art.

[0011] A first aspect of this application provides a human posture data privacy protection management system based on layered encryption and dynamic access control, including a monitoring device and a user client connected to a data management platform server, wherein: The monitoring device is configured to collect raw human posture data and send it to the data management platform server; The data management platform server is configured as follows: The system receives the raw human posture data and performs multi-dimensional privacy-sensitive quantification on the data, dynamically dividing it into multiple data layers with different sensitivity levels based on the quantification results. Different access control policies are formulated for each data layer, and corresponding encryption operations are performed on the data of each layer. Among them, the data layer with the highest sensitivity is encrypted using an attribute-based encryption algorithm, and the identification information of the ward is embedded in its access policy; the first access control policy is used for other data layers except for the data layer with the highest sensitivity. Performing identity verification and authorization operations based on access requests specifically includes: responding to data access requests from user clients, determining the sensitivity level of the requested data, and if the request is to access the data layer with the highest sensitivity, forcibly initiating a strong authentication process for the user's identity. After successful authentication, a temporary access credential is generated, and the temporary access credential and ciphertext generated by an attribute-based encryption algorithm are sent to the user client. For requests to access other data layers, verification is performed based on the first access control policy, and after successful verification, the corresponding protected data is sent to the user client. The user client is configured as follows: Send a data access request to the data management platform server; After requesting access to the most sensitive data layer and passing strong authentication, the system receives the temporary access credential and ciphertext generated by the attribute-based encryption algorithm from the server, and decrypts the ciphertext in a local secure environment based on the temporary access credential and the user's private key stored locally. When requesting access to other data layers, the protected data is received from the server based on the first access control policy and then decrypted or displayed.

[0012] In some embodiments of this application, the data management platform server is further configured as follows: The system records the identity verification operation, access requests and authorization events for all sensitive data levels, and generates an immutable audit log. For access to the highly sensitive data, the audit log records at least the success or failure status of the identity verification operation and the issuance event of temporary access credentials.

[0013] In some embodiments of this application, performing multi-dimensional privacy-sensitive metric operations on the data includes: The identification risk score is determined based on a predefined joint weight table, wherein the joint weight table classifies at least the hip joint, knee joint, ankle joint and pelvic joint as high-risk joints and assigns them a weight value of 0.9 to 1.0. Behavior categories are identified based on a lightweight temporal convolutional network model, and the behavior categories are mapped to behavior sensitivity scores according to a predefined behavior sensitivity dictionary, wherein the behavior sensitivity dictionary assigns a sensitivity score of 0.9 to 1.0 to behaviors such as falling and going to the toilet.

[0014] In some embodiments of this application, the step of dynamically dividing the data into multiple data layers with different sensitivity levels based on the quantization results includes: If the identity recognition risk score reaches 0.7 or higher, or the behavior sensitivity score reaches 0.8 or higher, then the corresponding data will be assigned to the data layer with the highest sensitivity.

[0015] In some embodiments of this application, the lightweight temporal convolutional network model includes at least four one-dimensional causal temporal convolutional layers, each with a kernel size of 3, and dilation coefficients set to 1, 2, 4, and 8 from the bottom layer to the top layer.

[0016] A second aspect of this application provides a method for privacy protection management of human posture data based on layered encryption and dynamic access control, executed by the data management platform server, including: The system receives raw human posture data and dynamically divides it into multiple data layers with different sensitivity levels based on the multi-dimensional privacy-sensitive quantification results of the data. Different access control policies are formulated for each data layer, and corresponding encryption operations are performed on the data of each layer. Among them, the data layer with the highest sensitivity is encrypted using an attribute-based encryption algorithm, and the access policy of the data layer embeds the identification information of the person under guardianship. The first access control policy is used for the other data layers except for the data layer with the highest sensitivity. Performing identity verification and authorization operations based on access requests specifically includes: responding to data access requests from user clients, determining the sensitivity level of the requested data, and if the request is to access the data layer with the highest sensitivity, forcibly initiating a strong authentication process for the user's identity. After successful authentication, a temporary access credential is generated, and the temporary access credential and ciphertext generated by an attribute-based encryption algorithm are sent to the user client. If the request is to access other data layers, verification is performed based on the first access control policy, and after successful verification, the protected data is sent to the user client.

[0017] In some embodiments of this application, after receiving the raw human posture data, the process further includes: The original human posture data undergoes joint coordinate normalization preprocessing, including: The trunk reference length is calculated based on the coordinates of the cervical and pelvic joints, using the following formula: L ref =||P neck -P pelvis || , where P neck P represents the coordinate data of the cervical joint points. pelvis This contains coordinate data of pelvic joint points. The scaling factor is determined based on the ratio of the trunk reference length to the preset standard trunk length, and the coordinates of each joint point are scaled to eliminate differences in individual absolute position, orientation, and body shape.

[0018] A third aspect of this application provides a method for accessing human posture data based on hierarchical encryption and dynamic access control, executed by the user client, including: Send an access request for a specific data layer to the data management platform server; After authentication via the data management platform server, corresponding data is received from the server, wherein: If a request is made to access the most sensitive data layer, then after strong authentication triggered by the server, a temporary access credential issued by the server and ciphertext generated using an attribute-based encryption algorithm are received from the server. If a request is made to access other data layers, then after verification based on the first access control policy, the corresponding protected data is received from the server. The received data is processed to achieve local presentation. Specifically, the received ciphertext is synthesized into a decryption attribute set based on the temporary access credential in a local secure environment, and then decrypted using the user's private key stored locally before presentation. The received protected data is either decrypted or directly displayed.

[0019] In some embodiments of this application, the step of synthesizing a decryption attribute set based on the temporary access credential in a local secure environment and then decrypting it using the locally stored user private key includes: Verify the digital signature of the temporary access credential to ensure it was issued by a legitimate data management platform server; Extract the session information for this authorization from the verified temporary access credential, including the user identifier and the access target identifier; Synthesize a decrypted attribute set that includes the user identifier, role, and access target identifier; After verifying that the decryption attribute set satisfies the access policy of the ciphertext by calling the local attribute-based decryption algorithm, the decryption operation is performed to obtain the plaintext data.

[0020] In some embodiments of this application, the ciphertext is generated from the human posture data of the person under guardianship using an attribute-based encryption algorithm; The verification that the decryption attribute set satisfies the ciphertext access strategy includes: Verify whether the decryption attribute set satisfies the condition that the ward's number is consistent with the embedded number in the ciphertext.

[0021] In summary, the human posture data privacy protection management system and method based on layered encryption and dynamic access control provided in the embodiments of this application, through the synergistic effect of the data layering module and quantification method combined with the layered encryption / decryption and policy management modules, enables the system to apply high-strength encryption protection to highly sensitive data, while employing lightweight processing to preserve the utility of medium- and low-sensitivity data. This achieves the optimal balance between privacy protection strength and data availability, thus solving the problem of severely compromised data utility or data rigidity caused by excessive desensitization or overall encryption in traditional solutions, achieving the optimal balance between privacy protection and data utility. Through the linkage of the dynamic access control engine and the identity verification module, real-time and accurate adaptation of access permissions and security context (such as user identity and operation urgency) is achieved. This achieves synchronous optimization of permission control granularity and security performance, thus resolving the contradiction between rigid permission management and dynamic security requirements, greatly improving management efficiency and security. Furthermore, the audit log module records key operations in an immutable manner, and the system records key events throughout the entire process from identity verification and access requests to authorization and decryption in detail. This achieves traceability, verifiability, and auditability of the technical processing, thereby solving the compliance risks and insufficient credibility issues caused by the lack of auditability in existing technologies, and providing a solid technical foundation for meeting the requirements of regulations such as GDPR. Attached Figure Description

[0022] The features and advantages of this application will become clearer with reference to the accompanying drawings, which are illustrative and should not be construed as limiting the application in any way. In the drawings: Figure 1 This is a schematic diagram of the architecture of the human posture data privacy protection management system to which this application applies; Figure 2 This is a flowchart illustrating a method for protecting human posture data privacy based on hierarchical encryption and dynamic access control, according to some embodiments of this application. Figure 3 This is a schematic diagram illustrating the layered encryption strategy implemented in some embodiments of this application; Figure 4 This is a flowchart illustrating the authorized access and decryption process performed in some embodiments of this application; Figure 5 This is a flowchart illustrating a method for protecting human posture data privacy based on hierarchical encryption and dynamic access control, according to some embodiments of this application. Detailed Implementation

[0023] In the following detailed description, numerous specific details of this application are illustrated by example to provide a thorough understanding of the relevant disclosure. However, it will be apparent to those skilled in the art that this application can be practiced without these details. It should be understood that the terms “system,” “apparatus,” “unit,” and / or “module” used in this application are one way of distinguishing different parts, elements, sections, or components at different levels in a sequential arrangement. However, these terms may be replaced with other expressions if other expressions can achieve the same purpose.

[0024] Various structural diagrams are used in this application to illustrate various variations of the embodiments according to this application. It should be understood that the preceding or following structures are not intended to limit this application. The scope of protection of this application is determined by the claims.

[0025] As described in the background section, privacy protection of human posture data (such as skeletal key points and 3D point clouds) faces severe challenges in remote monitoring scenarios such as nursing homes and postpartum care centers. While directly transmitting or storing raw data ensures data integrity, it easily leads to privacy leaks and incurs high transmission latency and storage costs. Transmitting anonymized feature vectors, although balancing privacy and utility to some extent, still carries the risk of reverse derivation, and excessive anonymization severely sacrifices data usability. Furthermore, existing access control schemes mostly employ static permission models, which cannot be dynamically adjusted based on access context (such as user identity and urgency of operation), resulting in inaccurate permission binding, lack of emergency response, and difficulty in meeting the fine-grained security management needs of sensitive scenarios. More importantly, existing solutions generally lack auditable capabilities for the entire data access lifecycle, failing to record key operations such as identity verification, access requests, and authorization decisions in an immutable manner. This not only makes it difficult for the system to meet compliance requirements of regulations such as GDPR but also severely restricts the system's transparency and credibility.

[0026] To address the aforementioned shortcomings, this application proposes a privacy protection management method for human posture data based on layered encrypted storage and dynamic access control. The core innovation of this solution lies in constructing a triple technical architecture integrating "data layering, dynamic control, and end-to-end auditing." First, by introducing a data layering mechanism with multi-dimensional privacy sensitivity quantification, the system can dynamically classify data into different security levels based on its inherent sensitivity and apply differentiated encryption protection strategies, thereby resolving the inherent contradiction between privacy protection and data utility at the source. Second, through a condition-triggered dynamic access control mechanism, the system can automatically adjust permission granting policies based on real-time security context, achieving a paradigm shift from "static authorization" to "dynamic adaptation." Finally, by establishing an audit trail system covering the entire operational chain, every key operational step is ensured to be traceable and verifiable, providing a solid technical foundation for system compliance.

[0027] Figure 1 This is a schematic diagram of the architecture of the human posture data privacy protection management system to which this application applies. Figure 1 As shown, the system includes monitoring equipment with data connectivity, a data management platform server, and user clients. Monitoring equipment deployed in nursing homes, postpartum care centers, and other monitoring scenarios collects raw human posture data and uploads it to the data management platform server. The server performs layered encryption and dynamic access control on the data based on multi-dimensional privacy-sensitive quantification results. Authorized users initiate requests to the server through the client, and after authorization verification, obtain the corresponding encrypted or protected data, and decrypt or display it locally.

[0028] The monitoring equipment may be depth cameras, millimeter-wave radar, or other monitoring devices deployed in nursing homes, postpartum care centers, or similar settings to acquire human posture data of the monitored individuals. The monitoring equipment collects raw human posture data and sends it to a data management platform server.

[0029] In some embodiments of this application, the data management platform server adopts a layered design, including: The data layering module is configured to perform multi-dimensional privacy-sensitive quantification operations on the raw data.

[0030] It includes: a sensitivity measurement submodule and a behavior recognition submodule.

[0031] The sensitivity quantification submodule is configured to determine the identity recognition risk score based on a predefined key point weight table, wherein the weight of high-risk key points is 0.9-1.0.

[0032] The behavior recognition submodule identifies behavior categories based on a lightweight temporal convolutional network model. The model contains at least four one-dimensional causal temporal convolutional layers, with each layer having a kernel size of 3 and dilation coefficients set to 1, 2, 4, and 8 from the bottom to the top layer, respectively. The behavior recognition submodule then maps the identified behavior categories to behavior sensitivity scores according to a predefined behavior sensitivity dictionary. The dictionary assigns a sensitivity score of 0.9 to 1.0 to behaviors such as falling and using the toilet, thereby providing a core quantitative basis for dynamic data stratification.

[0033] The policy management module can be configured to set different access control policies for each data layer in order to achieve fine-grained management through differentiated policies.

[0034] The layered encryption / decryption module is configured to implement differentiated encryption strategies for data layers with different sensitivities. For example, an attribute-based encryption algorithm is used for highly sensitive layers, and the access policy embeds the ward's identifier.

[0035] The dynamic access control engine is configured to perform identity verification and authorization operations in response to user requests through condition-triggered authentication.

[0036] The audit log module is configured to record critical operations such as identity verification and access requests, generating tamper-proof audit logs to meet compliance requirements of regulations such as GDPR.

[0037] The user client is configured with a secure environment to send user data access requests to the server and implement differentiated processing strategies for received data based on sensitivity levels. For example, for highly sensitive data, the encrypted data is decrypted based on temporary credentials and the user's private key.

[0038] This system constructs a complete data security protection system through the collaborative work of the aforementioned components. Its core design philosophy is: to achieve differentiated protection through layered processing, to adapt to complex scenarios through dynamic control, and to ensure compliance and trustworthiness through end-to-end auditing. The collaborative work of these components achieves an optimal balance between privacy protection, data utility, and system performance.

[0039] The following, in conjunction with the accompanying drawings, will explain in detail how the data management platform server and the user client implement the human posture data privacy protection method based on layered encryption and dynamic access control.

[0040] Figure 2 This is a flowchart illustrating a method for protecting human posture data privacy based on hierarchical encryption and dynamic access control, according to some embodiments of this application. In some embodiments, the method for protecting human posture data privacy based on hierarchical encryption and dynamic access control is comprised of... Figure 1 The data management platform server execution includes the following steps: S210: Receive raw human posture data and dynamically divide the data into multiple data layers with different sensitivity levels based on the multi-dimensional privacy-sensitive quantification results of the data.

[0041] Raw human posture data acquired from monitoring equipment such as depth cameras and millimeter-wave radar are typically represented as a time-series set of three-dimensional joint coordinates P={J}. i (t)=(x i (t),y i (t),z i (t))∣i=1,...,N;t=1,...,T}, where i is the joint number, such as i=1 representing the nose, N is the number of joints (such as 17 or 25), and T is the number of time frames.

[0042] First, the raw data is preprocessed, including the following steps: Coordinate normalization: With human pelvic joint Ppelvis Let x be the origin of space, and let the coordinates of all joint points be P(i) = (x... i ,y i ,z i The coordinates are converted to relative coordinates and then normalized based on the torso length to eliminate the influence of individual absolute position, initial orientation and body size differences.

[0043] 1. Calculate the reference length: Select the cervical joint point P. neck With P pelvis Calculate the reference length of the torso: L ref =||P neck -P pelvis || 2. Scale normalization: Set a standard torso length L. standard (e.g. L) standard =1.0), calculate the scaling factor (scale_factor), and scale the joint coordinates accordingly: scale_factor=L standard / L ref P normalized (i)=(P(i)-P pelvis )*scale_factor Among them, P normalized This is the final normalized joint coordinates that eliminate differences in position, orientation, and size.

[0044] Filtering and noise reduction: To address the continuous, low-frequency changes in human posture over time, two methods—standard Kalman filtering and low-pass filtering—are employed to smooth the data and effectively suppress noise introduced by sensor jitter. The standard Kalman filtering algorithm is used for smoothing joint coordinates: the process noise covariance matrix Q is set to diag(0.01, 0.01, 0.01) to accommodate the slow changes in human posture data; the observation noise covariance matrix R is set to diag(0.1, 0.1, 0.1). This parameter combination ensures data smoothing while accurately tracking the reasonable dynamic changes in human posture. The low-pass filtering uses a 5th-order low-pass filter with a cutoff frequency of 5Hz, which efficiently filters out high-frequency jitter noise while preserving the main frequency domain characteristics of human motion. Feature calculation: Extract static and dynamic features from the coordinate sequence, such as relative distance between joints, bone length, joint angle, motion speed, acceleration, etc., as input for subsequent sensitivity analysis.

[0045] Then, multi-dimensional privacy-sensitive quantification is performed on human posture features. Based on normalized coordinates that eliminate differences in individual position, orientation, and body size, an identity recognition risk score S is calculated. id (i) and behavioral sensitivity score S behavior (Data) provides a precise basis for subsequent data stratification. Specifically, it includes: 1) Identity verification risk score S id (i): This score is used to quantify a specific key point J. i The potential contribution of motion trajectories to identifying individual identities. The introduction of normalized coordinates allows for meaningful comparisons of keypoint data from different individuals within a unified scale space, thereby making prior knowledge derived from large-scale dataset analysis universal.

[0046] To minimize the computational overhead of the system, this application does not perform real-time model inference, but instead uses predefined rules based on the aforementioned prior knowledge to determine S. id The value of (i). Refer to the following example weight table for specific assignment: Joint type Example key points <![CDATA[Predefined weight S id > in accordance with High risk Hip, knee, ankle, pelvis 0.9-1.0 Individual unique Medium risk Shoulder width, elbows, chest cavity 0.4-0.6 Only has a certain degree of differentiation Low risk wrist, hand, foot, head 0.1-0.3 Trajectory changes By using this domain knowledge-based predefined approach, the system can obtain the risk score of key points simply by looking up a table when processing data in real time, completely avoiding complex online model calculations and greatly improving real-time performance.

[0047] 2) Behavioral sensitivity score S behavior (Data): This score is used to evaluate a specific data segment Data (consisting of normalized coordinates from multiple consecutive frames, Data={P}). normalized The degree of privacy sensitivity of the behavior described by (i,t)}. The specific calculation process is as follows: Based on data computation of dynamic features (such as speed and acceleration), and employing a lightweight Temporal Convolutional Network (TCN) architecture with a 1D-CNN structure for behavior recognition, this model achieves efficient capture of long sequence dependencies while significantly reducing computational complexity and parameter size through causal convolution and dilation mechanisms. This makes it more suitable for the efficiency requirements of real-time privacy-preserving scenarios. The model architecture design balances recognition accuracy and computational efficiency; the specific structure and hyperparameters are as follows: Input dimension description: The model input is a three-dimensional tensor [batch size, time steps T, feature dimension D]. The feature dimension D is determined by the product of the number of joints N and the dimension of the dynamic features (such as velocity, acceleration, etc.). For example, when there are 25 joints and each joint calculates 3-dimensional velocity + 3-dimensional acceleration, D = 25 × (3 + 3) = 150.

[0048] Core Network Structure Parameter Table Structural modules Specific parameter settings Design Description Number of convolutional layers 4-layer causal temporal convolutional layer Causal convolution ensures the rationality of temporal logic and prevents future information leakage. kernel size kernel_size=3 (uniform for each layer) Small convolutional kernels balance feature extraction capability and computational cost. expansion coefficient Set each layer as 1, 2, 4, 8. Exponentially expands the receptive field, efficiently capturing long-term temporal dependencies. Number of output channels The numbers increase progressively to 32, 64, 128, and 256. Step-by-step increase of feature dimensions to enhance non-linear feature representation capabilities Step length stride=1 Maintain the integrity of time-series data and avoid loss of critical motion information. Activation function ReLU (applied after each convolutional layer) Introducing nonlinearity enhances the model's ability to fit complex behavioral features. Regularization mechanism Dropout layer (drop rate = 0.2) This approach, which involves feeding the model after each convolutional layer, effectively suppresses overfitting and improves the model's generalization performance. To verify S behavior To verify the reliability of the (Data) mapping, this experiment conducted performance validation in a test environment using a fusion of publicly available datasets (Human3.6M). This dataset contains 72 hours of continuous posture data from 200 subjects, covering various typical behaviors under monitoring conditions. Test results show that the TCN model achieves an overall behavior recognition accuracy of 94.7%. Specifically, the accuracy rates for highly sensitive behaviors (falling, using the toilet, and violent struggling) are 96.3%, 95.8%, and 97.1%, respectively, while the accuracy rates for medium- and low-sensitivity behaviors (walking, sitting still, eating, etc.) are all no less than 93.2%. After 5-fold cross-validation, the model's average accuracy is 94.2%, with a standard deviation of only 0.8%, demonstrating excellent stability. These test results fully demonstrate that the TCN model can accurately and stably recognize various target behaviors, laying the foundation for subsequent S... behavior Accurate mapping of (Data) provides reliable label support.

[0049] 3) After the TCN model outputs the final behavior labels (such as falling, using the toilet, walking, sitting still, etc.), it completes the score mapping by querying a predefined behavior sensitivity dictionary. This dictionary is formulated based on the privacy protection needs of the guardianship scenario, dividing behaviors into different sensitivity levels according to "personal safety relevance" and "privacy exposure degree," with each level corresponding to a fixed score range, as shown in the following example: Behavior Categories <![CDATA[Predefined sensitivity S behavior > illustrate Falling down, going to the toilet 0.9-1.0 Involving personal safety and a high degree of privacy violent struggle 0.8-0.9 Represents an emergency Normal walking, sitting 0.1-0.3 Daily public behavior S of one segment of data behavior (Data) is determined by the highest score in the main behavioral category it contains.

[0050] Finally, based on the aforementioned sensitivity metrics scores, the data is dynamically stratified, and the permissions of family members over the ward are clarified based on these stratifications. Specifically, this includes the following stratifications: Meta-event and statistics layer (low sensitivity), behavioral summary and anonymized trajectory layer (medium sensitivity) and Raw data and a high-sensitivity behavior layer (high sensitivity).

[0051] S220, different access control policies are formulated for each data layer, and corresponding encryption operations are performed on the data of each layer. Among them, the data layer with the highest sensitivity is encrypted using an attribute-based encryption algorithm, and the access policy of the data layer embeds the identification information of the person under guardianship. The first access control policy is used for other data layers except for the data layer with the highest sensitivity.

[0052] This step serves as a bridge between data layering and dynamic access control. Its core is to formulate precise access policies for each data layer based on the quantization score calculated by s210 and to perform encryption.

[0053] Definition of hierarchical access strategy based on quantified scores The strategy management module calculates the identity verification risk score S based on the results of step one. id (i) and behavioral sensitivity score S behavior (Data), and based on the "high sensitivity first" logical principle (that is, if the score of any dimension reaches the threshold of the higher level, the data is assigned to the higher level), the data is divided into three logical levels, and the corresponding relationship is shown in the table below: The policy management module generates access policies for each data layer. To ensure that data permissions are precisely linked to specific wards, the policy embeds the ward's ID (e.g., J0001) as a core attribute.

[0054] Medium to low sensitivity layer: A static control strategy is adopted, allowing access only when the role attributes meet the requirements, such as Role = Caregiver. Highly sensitive layer: Access Policy: Access to this layer is subject to the strictest control. Except for administrators who may access it when necessary (such as for auditing), access permissions for family members are strongly tied to "identity verification." Only after successful verification can family members temporarily and selectively access the data of their corresponding ward at this layer, in response to urgent or special circumstances requiring in-depth verification.

[0055] The high-sensitivity layer employs an attribute-based encryption (ABE) algorithm to achieve a cryptographic binding between data and access policies. Specifically, this includes: When it is necessary to encrypt the highly sensitive plaintext data M of the ward J0001, the data is first bound to the number J0001. Then, the module confirms that the number is correct and sets the access policy A as follows: A=(Role=Administrator)||((Role=Family Member)&&(UserID=U0001)&&(Access Target=J0001)) Subsequently, the encryption module performs the operation CT = ABE.Encrypt(PK, M, A), where PK is the system public key. This operation combines policy A with data attribute A.data Both are encoded into the ciphertext CT. Therefore, decryption will only succeed if the user-provided attribute set S satisfies A(S)=1 and the user possesses the private key generated during registration based on the attributes and password. This especially requires access to the target U. target It must be consistent with the data attribute A embedded in the ciphertext. data The .belongs_to (i.e., JO001) function performs an exact match, thus fundamentally defining the boundaries of data access in cryptography.

[0056] Figure 3 This is a schematic diagram illustrating the layered encryption strategy implemented in some embodiments of this application.

[0057] S230, perform identity verification and authorization operations based on access requests, specifically including: responding to the user client's data access request, determining the sensitivity level of the requested data, if the request is to access the data layer with the highest sensitivity, then forcibly initiating a strong authentication process for the user's identity, generating a temporary access credential after successful authentication, and sending the temporary access credential and ciphertext generated by the attribute-based encryption algorithm to the user client; if the request is to access other data layers, then verifying based on the first access control policy, and sending the protected data to the user client after successful verification.

[0058] Users need to register and initialize before accessing data, including: When a family member registers on the platform for the first time, the platform generates a unique user ID (e.g., U0001) after the kinship relationship is manually confirmed, and automatically generates a ward number to bind to the account. At this point, the account has completed the binding of the user ID, kinship relationship, and ward number. Simultaneously, the system generates a unique user private key (SK) on the user's local client based on their attribute set (such as role, bound ward number, etc.) and user password. When the user changes devices or their attributes (such as adding a bound ward) change, the system will regenerate the public and private keys, with the new public key automatically uploaded and updated to the server.

[0059] For regular user logins, only username and password are required for subsequent logins. Strong identity verification is only required when adding or removing a ward from a ward, or when performing sensitive actions such as adding or removing a ward from a ward.

[0060] When a user requests access to data, the dynamic access control engine, in conjunction with the identity verification module, executes a fine-grained, condition-triggered privilege escalation process. The core of this process lies in dynamically generating temporary, high-privilege access credentials for legitimate users through strong authentication. This includes: Request parsing and basic permission verification: User access automatically retrieves the kinship relationship with the ward. The engine first verifies whether the user's static role attribute (family member) has the normal permissions to access the data with that ID. This step ensures that the request is logically valid.

[0061] Figure 4 This is a flowchart illustrating the authorized access and decryption process performed in some embodiments of this application. For example... Figure 4 As shown: The system determines the sensitivity level of the requested data and adopts a differentiated processing strategy based on different levels. Access to the highly sensitive data layer: When a user requests access to raw data and highly sensitive behavior layers, the system forcibly initiates an identity verification process. This automatically retrieves the user ID and verifies whether the attributes match.

[0062] After successful verification, the server confirms the user's identity and permissions are correct, and then generates an access token containing authorization information for this session (such as user ID, role, and access target ID). This token is digitally signed by the server to prove the user's access rights to specific data in this session. The system then sends the access token and the encrypted data (CT) requested by the user to the user's client via a TLS / SSL encrypted channel.

[0063] Regular access to non-sensitive data layers: For access requests to the meta-event layer and behavior summary layer, the system will not trigger strong authentication. The ABE decryption algorithm directly uses the user's static attribute set, such as ({role=family member, access target=O001}), to match the relatively loose encryption strategy of the corresponding data layer, thus achieving conventional on-demand decryption.

[0064] In some embodiments of this application, all access requests and identity verification operations (whether successful or not) are recorded in detail to form an immutable audit log.

[0065] Figure 5 This is a flowchart illustrating a method for protecting human posture data privacy based on hierarchical encryption and dynamic access control, according to some embodiments of this application. In some embodiments, the method for protecting human posture data privacy based on hierarchical encryption and dynamic access control is comprised of... Figure 1 The user client execution specifically includes the following steps: S510 sends an access request for a specific data layer to the data management platform server.

[0066] In some embodiments of this application, sending a needle access request to the data management platform server requires specifying the data layer identifier (such as a high-sensitivity layer, a medium-sensitivity layer, or a low-sensitivity layer) and the target ward's number.

[0067] S520, after authentication through the data management platform server, corresponding data is received from the server, wherein: if the request is to access the most sensitive data layer, after strong authentication triggered by the server, a temporary access credential issued by the server and ciphertext generated by an attribute-based encryption algorithm are received from the server; if the request is to access other data layers, after verification based on the first access control policy, the corresponding protected data is received from the server.

[0068] In some embodiments of this application, if a request is made to access the most sensitive data layer, after strong authentication triggered by the server, the security module receives a temporary access credential issued by the server and ciphertext generated using an attribute-based encryption algorithm on the client's local machine.

[0069] If a request is made to access other data layers, the client will receive the lightly encrypted or de-identified data directly from the server after authentication through a role-based access control policy.

[0070] S530, the received data is processed to achieve local presentation. Specifically, the received ciphertext is synthesized into a decryption attribute set based on the temporary access credential in a local secure environment, and then decrypted using the locally stored user private key before presentation. The received protected data is either decrypted or directly displayed.

[0071] When high-sensitivity data is received, the digital signature of the temporary access credential is first verified in the local security environment. After confirming its legitimacy, the session authorization information is extracted, and then a decryption attribute set containing user identifier, role permissions, and access target identifier is synthesized. Subsequently, an attribute-based decryption algorithm is called to perform matching verification and decryption. The entire process is completed in a hardware-level isolated environment to ensure security. For medium- and low-sensitivity data, a simplified processing flow is adopted. After verifying the data integrity, symmetric decryption or parsing and display are performed directly. This layered processing mechanism not only ensures the security requirements of high-sensitivity data, but also optimizes the processing efficiency of regular data.

[0072] In practice, the client stores the user's private key in a hardware-level secure area and performs decryption operations to ensure that sensitive information is always protected. For highly sensitive ciphertext, it is necessary to strictly match the consistency conditions between the decryption attribute set and the access policy, including verifying the correspondence between the guardian's number and the embedded identifier in the ciphertext. For medium and low sensitive data that has undergone lightweight encryption or desensitization, it is directly and quickly decrypted or visualized according to preset permissions. This design effectively balances the dual requirements of system security and real-time response.

[0073] To verify the comprehensive advantages of the method described in this application (hereinafter referred to as "this method") in terms of data utility and processing efficiency, one embodiment of this application selects two mainstream methods as control groups for comparative experiments with this method: Method A (directly storing / transmitting raw posture data) and Method B (transmitting only desensitized feature vectors), with this method serving as the experimental group.

[0074] Experimental environment The study included 200 participants (100 elderly people in nursing homes and 100 mothers and infants in postpartum care centers), covering different ages, genders and behavioral characteristics; each participant collected 72 hours of continuous posture data, with a sampling frequency of 30Hz, and each frame of data contained 3D coordinates of 25 joints (approximately 1.2KB / frame).

[0075] The total amount of raw data collected was approximately 1800GB, of which the amount of data from a single person was approximately 9GB. Based on the hierarchical storage strategy of this method, the meta-event and statistical layer (low sensitivity) accounts for 5% of the total raw data, with a storage capacity of 90GB (0.1KB per data entry, recording statistical information such as behavior category and occurrence time); the behavior summary and anonymized trajectory layer (medium sensitivity) accounts for 15% of the total raw data, with a storage capacity of 270GB (0.3KB per data entry, containing anonymized keypoint trajectories and dynamic features); and the raw data and high-sensitivity behavior layer (high sensitivity) accounts for 30% of the total raw data, with a storage capacity of 540GB (only the raw frame data corresponding to high-sensitivity behaviors are retained).

[0076] Specific test results Evaluation indicators Method A (Raw Data) Method (Desensitization Feature Vector) This method Storage usage 1800GB 720GB 900GB Transmission delay (per frame) 156ms 89ms 63ms Behavior recognition accuracy 98.1% 78.2% 94.2% Results Analysis Balancing Data Utility and Efficiency: While traditional de-identification methods reduce storage costs, excessive de-identification leads to a decrease in accuracy. This solution significantly reduces storage requirements while maintaining accuracy at only 3.9% lower than direct data transmission, achieving an optimal balance between privacy protection and data utility.

[0077] Advantages of the tiered strategy: By using tiered storage, only 50% of the original data volume is needed to meet the requirements of high recognition rate behavior recognition, which verifies the efficiency of the storage strategy.

[0078] In summary, the human posture data privacy protection management system and method based on layered encryption and dynamic access control provided in the embodiments of this application, through the synergistic effect of the data layering module and quantification method combined with the layered encryption / decryption and policy management modules, enables the system to apply high-strength encryption protection to highly sensitive data, while employing lightweight processing to preserve the utility of medium- and low-sensitivity data. This achieves the optimal balance between privacy protection strength and data availability, thus solving the problem of severely compromised data utility or data rigidity caused by excessive desensitization or overall encryption in traditional solutions, achieving the optimal balance between privacy protection and data utility. Through the linkage of the dynamic access control engine and the identity verification module, real-time and accurate adaptation of access permissions and security context (such as user identity and operation urgency) is achieved. This achieves synchronous optimization of permission control granularity and security performance, thus resolving the contradiction between rigid permission management and dynamic security requirements, greatly improving management efficiency and security. Furthermore, the audit log module records key operations in an immutable manner, and the system records key events throughout the entire process from identity verification and access requests to authorization and decryption in detail. This achieves traceability, verifiability, and auditability of the technical processing, thereby solving the compliance risks and insufficient credibility issues caused by the lack of auditability in existing technologies, and providing a solid technical foundation for meeting the requirements of regulations such as GDPR.

[0079] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the devices and modules described above can be referred to the corresponding descriptions in the foregoing device embodiments, and will not be repeated here.

Claims

1. A human posture data privacy protection management system based on hierarchical encryption and dynamic access control, characterized in that, This includes monitoring equipment and user clients that are connected to the data management platform server, among which: The monitoring device is configured to collect raw human posture data and send it to the data management platform server; The data management platform server is configured as follows: The system receives the raw human posture data and performs multi-dimensional privacy-sensitive quantification on the data, dynamically dividing it into multiple data layers with different sensitivity levels based on the quantification results. Different access control policies are formulated for each data layer, and corresponding encryption operations are performed on the data of each layer. Among them, the data layer with the highest sensitivity is encrypted using an attribute-based encryption algorithm, and the identification information of the ward is embedded in its access policy; the first access control policy is used for other data layers except for the data layer with the highest sensitivity. Performing identity verification and authorization operations based on access requests specifically includes: responding to data access requests from user clients, determining the sensitivity level of the requested data, and if the request is to access the data layer with the highest sensitivity, forcibly initiating a strong authentication process for the user's identity. After successful authentication, a temporary access credential is generated, and the temporary access credential and ciphertext generated by an attribute-based encryption algorithm are sent to the user client. For requests to access other data layers, verification is performed based on the first access control policy, and after successful verification, the corresponding protected data is sent to the user client. The user client is configured as follows: Send a data access request to the data management platform server; After requesting access to the most sensitive data layer and passing strong authentication, the system receives the temporary access credential and ciphertext generated by the attribute-based encryption algorithm from the server, and decrypts the ciphertext in a local secure environment based on the temporary access credential and the user's private key stored locally. When requesting access to other data layers, the protected data is received from the server based on the first access control policy and then decrypted or displayed.

2. The system according to claim 1, characterized in that, The data management platform server is also configured as follows: The system records the identity verification operation, access requests and authorization events for all sensitive data levels, and generates an immutable audit log. For access to the highly sensitive data, the audit log records at least the success or failure status of the identity verification operation and the issuance event of temporary access credentials.

3. The system according to claim 1, characterized in that, The multi-dimensional privacy-sensitive quantification operation performed on the data includes: The identification risk score is determined based on a predefined joint weight table, wherein the joint weight table classifies at least the hip joint, knee joint, ankle joint and pelvic joint as high-risk joints and assigns them a weight value of 0.9 to 1.

0. Behavior categories are identified based on a lightweight temporal convolutional network model, and the behavior categories are mapped to behavior sensitivity scores according to a predefined behavior sensitivity dictionary, wherein the behavior sensitivity dictionary assigns a sensitivity score of 0.9 to 1.0 to behaviors such as falling and going to the toilet.

4. The system according to claim 3, characterized in that, The dynamic division of the data into multiple data layers with different sensitivity levels based on the quantization results includes: If the identity recognition risk score reaches 0.7 or higher, or the behavior sensitivity score reaches 0.8 or higher, then the corresponding data will be assigned to the data layer with the highest sensitivity.

5. The system according to claim 1, characterized in that: The lightweight temporal convolutional network model contains at least four one-dimensional causal temporal convolutional layers, with each layer having a kernel size of 3 and dilation coefficients set to 1, 2, 4, and 8 from the bottom layer to the top layer.

6. A method for protecting and managing human posture data privacy based on layered encryption and dynamic access control, executed by the data management platform server, characterized in that, include: The system receives raw human posture data and dynamically divides it into multiple data layers with different sensitivity levels based on the multi-dimensional privacy-sensitive quantification results of the data. Different access control policies are formulated for each data layer, and corresponding encryption operations are performed on the data of each layer. Among them, the data layer with the highest sensitivity is encrypted using an attribute-based encryption algorithm, and the access policy of the data layer embeds the identification information of the person under guardianship. The first access control policy is used for the other data layers except for the data layer with the highest sensitivity. Performing identity verification and authorization operations based on access requests specifically includes: responding to data access requests from user clients, determining the sensitivity level of the requested data, and if the request is to access the data layer with the highest sensitivity, forcibly initiating a strong authentication process for the user's identity. After successful authentication, a temporary access credential is generated, and the temporary access credential and ciphertext generated by an attribute-based encryption algorithm are sent to the user client. If the request is to access other data layers, verification is performed based on the first access control policy, and after successful verification, the protected data is sent to the user client.

7. The method according to claim 6, characterized in that, After receiving the raw human posture data, the process also includes: The original human posture data undergoes joint coordinate normalization preprocessing, including: The trunk reference length is calculated based on the coordinates of the cervical and pelvic joints, using the following formula: L ref =||P neck -P pelvis ||, Among them, P neck P represents the coordinate data of the cervical joint points. pelvis This contains coordinate data of pelvic joint points. The scaling factor is determined based on the ratio of the trunk reference length to the preset standard trunk length, and the coordinates of each joint point are scaled to eliminate differences in individual absolute position, orientation, and body shape.

8. A method for accessing human posture data based on hierarchical encryption and dynamic access control, executed by the user client, characterized in that, include: Send an access request for a specific data layer to the data management platform server; After authentication via the data management platform server, corresponding data is received from the server, wherein: If a request is made to access the most sensitive data layer, then after strong authentication triggered by the server, a temporary access credential issued by the server and ciphertext generated using an attribute-based encryption algorithm are received from the server. If a request is made to access other data layers, then after verification based on the first access control policy, the corresponding protected data is received from the server. The received data is processed to achieve local presentation. Specifically, the received ciphertext is synthesized into a decryption attribute set based on the temporary access credential in a local secure environment, and then decrypted using the user's private key stored locally before presentation. The received protected data is either decrypted or directly displayed.

9. The method according to claim 8, characterized in that, The process of synthesizing a decryption attribute set based on the temporary access credential in the local secure environment, and then decrypting it using the locally stored user private key, includes: Verify the digital signature of the temporary access credential to ensure it was issued by a legitimate data management platform server; Extract the session information for this authorization from the verified temporary access credential, including the user identifier and the access target identifier; Synthesize a decrypted attribute set that includes the user identifier, role, and access target identifier; After verifying that the decryption attribute set satisfies the access policy of the ciphertext by calling the local attribute-based decryption algorithm, the decryption operation is performed to obtain the plaintext data.

10. The method according to claim 9, characterized in that: The ciphertext is generated from the ward's body posture data using an attribute-based encryption algorithm; The verification that the decryption attribute set satisfies the ciphertext access strategy includes: Verify whether the decryption attribute set satisfies the condition that the ward's number is consistent with the embedded number in the ciphertext.

Images