Intelligent door lock dynamic authorization control method and system based on context awareness

Abstract

The invention relates to an intelligent door lock dynamic authorization control method and system based on context awareness, and the method comprises the steps: triggering a multi-dimensional context information collection instruction when an access request without a preset permission is detected, obtaining multi-dimensional context information corresponding to the current access request, and forming multi-dimensional context data; preprocessing the multi-dimensional situation data to obtain a multi-dimensional situation vector; inputting the multi-dimensional situation vector into a pre-trained dynamic risk assessment model, and calculating a trust score of the access request; automatically executing an adaptive authorization decision according to the interval to which the trust score belongs; and recording the situation information, the calculation process, the interaction content and the authorization result of the whole process of the access request, and generating a traceable authorization log of a time axis structure. The method and the device have the effect of improving user security.

Description

Technical Field

[0001] This application relates to the technical field of smart door locks, and in particular to a dynamic authorization control method and system for smart door locks based on context awareness. Background Technology

[0002] Currently, smart locks, as an important terminal for the integration of the Internet of Things (IoT) and home security, are widely used in residential, apartment, and office areas. Traditional smart lock control mechanisms typically rely on fixed authorization methods, such as password input, fingerprint recognition, or temporary authorization via mobile devices. However, in actual use, if the visitor has not registered authorization information in advance, remote user intervention or manual assessment of access security is required. This not only increases the user's operational burden but also fails to adequately handle complex access scenarios.

[0003] While some existing smart access control systems support remote authorization via mobile devices, most lack the ability to analyze access behavior in real time, making it difficult to dynamically determine access requests based on factors such as time, environment, and identity characteristics. For example, late at night or when the user is not locally, the system may still execute an unlocking operation based on a single identity match, posing a potential security risk.

[0004] The existing technical solutions mentioned above have the following drawbacks: Most existing smart door lock authorization control methods are static permission verification, which lack the ability to comprehensively judge the multi-dimensional contextual factors behind access requests and cannot dynamically adjust the authorization strategy based on real-time risks. Therefore, there are problems of insufficient security and poor user experience, and there is room for improvement. Summary of the Invention

[0005] To enhance user security, this application provides a context-aware smart lock dynamic authorization control method and system.

[0006] The above-mentioned objective of this application is achieved through the following technical solution: A context-aware dynamic authorization control method for smart locks, comprising: When an access request without preset permissions is detected, a multi-dimensional context information collection instruction is triggered to obtain multi-dimensional context information corresponding to the current access request, forming multi-dimensional context data. The multi-dimensional context information includes time context, environmental context, user data context, and location context. The multidimensional context data is preprocessed to obtain a multidimensional context vector; The multidimensional context vector is input into a pre-trained dynamic risk assessment model to calculate the trust score of the access request; Automatically execute adaptive authorization decisions based on the trust score range; Record the context information, calculation process, interaction content and authorization results of the entire access request process, and generate a traceable authorization log with a timeline structure.

[0007] By adopting the above technical solution, when an access request without preset permissions is detected, a multi-dimensional contextual information collection instruction is triggered to form multi-dimensional contextual data. This allows for the introduction of multi-dimensional information sources rather than single identity information at the moment the access request is generated, thereby significantly enhancing the completeness and accuracy of access determination. By preprocessing the multi-dimensional contextual data to obtain multi-dimensional contextual vectors, the original heterogeneous data can be converted into a standardized structured representation that can be used for model calculation, thereby improving the input quality and operational stability of the risk assessment model. By inputting the multi-dimensional contextual vectors into a pre-trained dynamic risk assessment model to calculate the trust score of the access request, real-time quantifiable evaluation of the access request's credibility can be achieved, thereby reducing manual intervention and improving the speed of access determination. By automatically executing adaptive authorization decisions based on the trust score range, the door lock can automatically adopt different levels of response strategies according to different risk levels, thereby ensuring security while reducing interference with normal access. By recording the contextual information, calculation process, interaction content, and authorization results of the entire access request process and generating a traceable authorization log with a timeline structure, the traceability of the entire access chain can be achieved, thereby supporting security auditing, anomaly review, and subsequent model optimization.

[0008] In one example, this application can be further configured as follows: when an access request without preset permissions is detected, a multi-dimensional context information collection instruction is triggered to obtain multi-dimensional context information corresponding to the current access request, forming multi-dimensional context data, specifically including: The system uses a door lock camera to capture images of the visit scene in real time, and analyzes these images based on an image recognition algorithm to extract visitor feature data in order to generate environmental context information. The visitor feature data includes visitor facial features, head orientation, occlusion behavior, and clothing type. The system collects ambient sound and visitor voice information at the site using a microphone, and performs keyword recognition and audio state analysis on the visitor voice information to supplement the environmental context information. Obtain the current access timestamp, decompose and analyze the access timestamp, and construct time context information in the form of time tags; By reading the user's calendar events and communication keywords through a secure communication interface with the user's mobile terminal, user data context information is generated; The system acquires GPS data from the user's mobile terminal and determines the user's geographical location status based on the GPS data, thereby forming location context information. The environmental context information, time context information, user data context information, and location context information are statistically integrated to form a multidimensional context vector.

[0009] By employing the aforementioned technical solutions, and by capturing images of the visitor's environment using cameras and extracting information such as facial features, head orientation, and occlusion behavior, the system can identify the visitor's true state and behavioral characteristics, thereby enhancing the accuracy of environmental context information. By collecting ambient sound and visitor speech and performing keyword recognition and audio analysis, the system can uncover the visitor's voice characteristics and possible visit purposes, supplementing contextual factors that images alone cannot identify. By decomposing visit timestamps and constructing time tags, the system can identify whether the current visit falls within the user's normal activity cycle, thus reducing the potential risk of visits during abnormal time periods. By reading user calendar events and communication keywords to obtain user data context information, the system can determine whether the visit request is related to the user's recent schedule, thereby improving the ability to detect the rationality of the visit purpose. By identifying the user's location status based on GPS data, the system can determine whether the user is near a door lock, thereby reducing the risk of impersonation when the user is far away. By integrating time context, environmental context, user data context, and location context, the system can generate complete, complementary, and multi-dimensional visit context information, thereby enabling subsequent risk assessment models to have higher semantic understanding capabilities and judgment reliability.

[0010] In one example, this application can be further configured such that: preprocessing the multidimensional context data to obtain a multidimensional context vector specifically includes: Feature extraction is performed on the multidimensional contextual data to extract multiple scoring factors used to characterize access credibility. These scoring factors include: Visitor identity tags are generated by comparing visitor facial features with a user-preset trusted facial image database. The time matching status is generated by comparing the current access time with the time period of the user's scheduled calendar event. The location proximity marker is generated based on the distance between the user's GPS data and the door lock location. The semantic matching score is generated based on the similarity calculation between visitor voice information and user communication keywords; Behavioral pattern matching scores are generated based on similarity analysis between the current access context and historical successful authorization behaviors. The scoring factors are structured and encoded according to preset feature dimensions to form a multidimensional context vector for model input.

[0011] By adopting the above technical solutions, multiple scoring factors, including identity markers, time matching status, location proximity markers, semantic matching scores, and behavioral pattern matching scores, can be extracted from multidimensional contextual data. This allows for fine-grained characterization of access credibility from different security perspectives, thereby effectively improving the judgment model's ability to distinguish complex access scenarios. Furthermore, by generating multidimensional contextual vectors through structured encoding of the scoring factors, the model's input data can be guaranteed to be in a uniform, stable, and reliable format, thereby improving the training quality and inference accuracy of the dynamic risk assessment model.

[0012] In one example, this application can be further configured as follows: the context-aware smart lock dynamic authorization control method further includes: A training dataset is constructed based on historical access request records. The training dataset includes a multidimensional context vector and access result labels formed by extracting and encoding multidimensional context information corresponding to each access request. The multidimensional context information includes time context, environmental context, user data context and location context. The access result labels include authorization success, authorization failure, user rejection, no response and abnormal alarm. An initial classification model is constructed using the random forest algorithm, and a supervised learning process is performed using the training dataset as input to optimize the model parameters, thereby obtaining the pre-trained dynamic risk assessment model. The pre-trained dynamic risk assessment model is deployed to a cloud server. When a multi-dimensional context vector generated in real time is received, a continuous trust score between 0 and 100 is output. The trust score is used to drive the execution logic of subsequent adaptive authorization decisions.

[0013] By adopting the above technical solutions, a training dataset containing context vectors and access result labels can be constructed based on historical access request records. This provides the model with training samples that cover real access behavior patterns, thereby improving the model's recognition ability and generalization performance in real-life scenarios. By using the random forest algorithm to construct a classification model and performing supervised learning to optimize parameters, the model's ability to mine multi-dimensional feature relationships can be enhanced, thus maintaining stable risk prediction performance when facing noise and abnormal data. By deploying the trained model to the cloud and outputting continuous trust scores, real-time, refined, and quantifiable credibility assessment of access requests can be achieved, thereby providing a reliable basis for adaptive authorization decisions and enhancing the intelligence capabilities of the door lock system.

[0014] In one example, this application can be further configured such that: the automatic execution of adaptive authorization decisions based on the trust score interval specifically includes: When the trust score is in the preset high confidence range, an automatic authorization operation is performed, directly sending an unlocking command to the door lock control module and simultaneously recording the authorization pass event; When the trust score is within the preset medium trust range, the semantic verification-assisted authorization process is triggered; When the trust score is in a preset low trust range, the current access request is rejected, a rejection prompt is displayed to the visitor, and a risk notification is pushed to the user's mobile terminal to alert them to potential abnormal access behavior.

[0015] By adopting the above technical solutions, automatic authorization can be directly executed when the trust score is in the high trust range, enabling seamless unlocking while ensuring security, thereby improving user convenience and experience continuity. By triggering semantic verification-assisted authorization when the trust score is in the medium trust range, the purpose of access can be further verified and some spoofing scenarios can be filtered out, thereby reducing the possibility of misauthorization due to medium-risk access. By denying access and pushing risk notifications to user terminals when the trust score is in the low trust range, potential malicious access behaviors can be blocked in a timely manner, thereby enhancing the system's protection capabilities and improving the user's response speed to security incidents.

[0016] In one example, this application can be further configured such that the triggering of the semantic verification-assisted authorization process specifically includes: When the trust score is in the preset medium trust range, a verification prompt voice is issued to the visitor, prompting the visitor to state the reason for the visit or to verify the password; Collect visitor response speech and perform speech recognition and natural language semantic analysis to extract keyword content from the response speech; The semantic similarity of the keyword content is matched with the communication keywords, calendar event descriptions or trust phrases set by the authorizing person extracted from the user's mobile terminal to obtain a semantic similarity matching score. When the semantic similarity matching score reaches the preset verification threshold, an auxiliary authorization instruction is executed to control the door lock to unlock; otherwise, the door lock remains closed and a notification is sent to the user to request confirmation.

[0017] By adopting the above technical solutions, when a visitor is in the medium-trust range, a verification prompt voice is initiated, requiring the visitor to state the reason for the visit or a verification password. This proactively confirms the visitor's intention, thereby improving the reliability of screening medium-risk visits. By recognizing and semantically analyzing the response voice and extracting keywords, the core content of the visitor's statement can be accurately captured, thereby enhancing the auxiliary value of language information in trust judgment. By semantically matching keywords with communication keywords, calendar events, or trust phrases and calculating scores, the degree of relevance between the visitor's statement and the user's real-life context can be quantitatively judged, thereby reducing the risk of impersonation. By performing auxiliary authorization when the similarity score reaches a threshold, security and usability can be balanced, thereby reducing the burden of frequent user intervention and maintaining a user-friendly experience for legitimate visitors.

[0018] In one example, this application can be further configured as follows: the context-aware smart lock dynamic authorization control method further includes: When abnormal access behavior is detected, the local sound and light alarm device is automatically triggered and a risk warning message is sent to the user's mobile terminal. At the same time, the current access record is marked as a high-risk event and included in the authorization log. The abnormal access behavior includes consecutive failed authorization attempts by the same identity within a short period of time, access time within a sensitive time period set by the user, and face occlusion or multiple people gathering in the image recognition result.

[0019] By adopting the above technical solutions, the automatic triggering of audible and visual alarms upon detecting abnormal access behavior can immediately deter potential intruders and reduce the probability of successful unauthorized access, thereby enhancing the proactive defense capabilities of the door lock system. Pushing risk alerts to users' mobile terminals allows users to be aware of security risks immediately, enabling them to take timely measures such as remote denial, alarm activation, or contacting security. Marking abnormal access records as high-risk events and including them in the authorization log provides crucial data support for subsequent security analysis, accountability, and further model training and optimization, thereby continuously improving the security and intelligence level of the entire smart door lock system.

[0020] The second objective of this invention is achieved through the following technical solution: A context-aware intelligent door lock dynamic authorization control system, comprising: The access triggering module is used to trigger a multi-dimensional context information collection instruction when an access request without preset permissions is detected, to obtain multi-dimensional context information corresponding to the current access request, and to form multi-dimensional context data. The multi-dimensional context information includes time context, environmental context, user data context, and location context. The data preprocessing module is used to preprocess the multidimensional context data to obtain a multidimensional context vector; The risk scoring module is used to input the multidimensional context vector into a pre-trained dynamic risk assessment model to calculate the trust score of the access request. The authorization decision module is used to automatically execute adaptive authorization decisions based on the range to which the trust score belongs; The logging module is used to record contextual information, calculation process, interaction content and authorization results of the entire access request process, and generate a traceable authorization log with a timeline structure.

[0021] By adopting the above technical solution, when an access request without preset permissions is detected, a multi-dimensional contextual information collection instruction is triggered to form multi-dimensional contextual data. This allows for the introduction of multi-dimensional information sources rather than single identity information at the moment the access request is generated, thereby significantly enhancing the completeness and accuracy of access determination. By preprocessing the multi-dimensional contextual data to obtain multi-dimensional contextual vectors, the original heterogeneous data can be converted into a standardized structured representation that can be used for model calculation, thereby improving the input quality and operational stability of the risk assessment model. By inputting the multi-dimensional contextual vectors into a pre-trained dynamic risk assessment model to calculate the trust score of the access request, real-time quantifiable evaluation of the access request's credibility can be achieved, thereby reducing manual intervention and improving the speed of access determination. By automatically executing adaptive authorization decisions based on the trust score range, the door lock can automatically adopt different levels of response strategies according to different risk levels, thereby ensuring security while reducing interference with normal access. By recording the contextual information, calculation process, interaction content, and authorization results of the entire access request process and generating a traceable authorization log with a timeline structure, the traceability of the entire access chain can be achieved, thereby supporting security auditing, anomaly review, and subsequent model optimization.

[0022] In summary, this application includes the following beneficial technical effects: 1. By triggering a multi-dimensional contextual information collection instruction and forming multi-dimensional contextual data when an access request without preset permissions is detected, multi-dimensional information sources can be introduced at the first moment when the access request is generated, rather than a single identity information, thereby significantly enhancing the completeness and accuracy of access determination. 2. By preprocessing the multidimensional context data to obtain multidimensional context vectors, the original heterogeneous data can be converted into a standardized structured representation that can be used for model calculation, thereby improving the input quality and model operation stability of the risk assessment model; by inputting the multidimensional context vectors into a pre-trained dynamic risk assessment model to calculate the trust score of the access request, real-time credible quantitative assessment of the access request can be achieved, thereby reducing manual intervention and improving the speed of access judgment. 3. By automatically executing adaptive authorization decisions based on the trust score range, the door lock can automatically adopt different response strategies according to different risk levels, thereby reducing interference with normal access while ensuring security; by recording the context information, calculation process, interaction content and authorization results of the entire access request process and generating a traceable authorization log with a timeline structure, the traceability of the entire access chain can be achieved, thereby supporting security auditing, anomaly review and subsequent model optimization. Attached Figure Description

[0023] Figure 1 This is a flowchart of a context-aware smart lock dynamic authorization control method according to an embodiment of this application; Figure 2 This is a flowchart illustrating the implementation of step S10 in a context-aware smart lock dynamic authorization control method according to an embodiment of this application. Figure 3 This is a flowchart illustrating the implementation of step S20 in a context-aware smart lock dynamic authorization control method according to an embodiment of this application. Figure 4 This is another implementation flowchart of step S30 in a context-aware smart lock dynamic authorization control method according to an embodiment of this application; Figure 5 This is a flowchart illustrating the implementation of step S40 in a context-aware smart lock dynamic authorization control method according to an embodiment of this application. Figure 6 This is a flowchart illustrating the implementation of step S42 in a context-aware smart lock dynamic authorization control method according to an embodiment of this application. Figure 7 This is another implementation flowchart of a context-aware smart lock dynamic authorization control method in one embodiment of this application; Figure 8 This is a principle block diagram of a context-aware intelligent door lock dynamic authorization control system according to one embodiment of this application. Detailed Implementation

[0024] The present application will be further described in detail below with reference to the accompanying drawings.

[0025] In one embodiment, such as Figure 1 As shown, this application discloses a context-aware dynamic authorization control method for smart door locks, which specifically includes the following steps: S10: When an access request without preset permissions is detected, a multi-dimensional context information collection instruction is triggered to obtain multi-dimensional context information corresponding to the current access request and form multi-dimensional context data. The multi-dimensional context information includes time context, environmental context, user data context and location context.

[0026] Specifically, upon receiving a request for action from an unauthorized or unregistered user at the door, the system immediately retrieves the door lock camera and microphone to initiate the data collection process. It then uses an image recognition engine to perform facial recognition, occlusion detection, head posture analysis, and clothing type classification on the captured images. Simultaneously, it performs voice separation and keyword extraction on the audio signal to supplement the environmental context dimension. The action logic synchronously records the current timestamp and converts it into a structured time tag, such as "weekday noon" or "weekend late night," to construct time context information. Furthermore, through a Bluetooth Low Energy connection or secure application interface established with the user's mobile phone, it retrieves the user's calendar entries and recent communication records, and compares the user's GPS data with the door lock location to calculate the real-time distance between the user and the door lock. This generates user data context and location context information. For example, if the user is far from the local area and the visitor at the door is a stranger, a preliminary risk indication is established. All collected information is ultimately integrated into a complete multi-dimensional contextual data structure.

[0027] S20: Preprocess the multidimensional context data to obtain a multidimensional context vector.

[0028] Specifically, the multidimensional contextual data is first standardized, including representing image features through vectorized encoding, such as face confidence scores and occlusion probabilities. Then, keywords in the speech content are matched and analyzed with communication records to generate semantic matching scores. Time data is mapped to numerical labels, such as "late night = 0.1" and "office day = 0.8". GPS distance is calculated to obtain physical distance quantification values ​​and mapped to spatial scores. Subsequently, all feature factors are encapsulated into a vector structure. For example, the scenario of "user away + late night + stranger + semantic mismatch" is encapsulated into a multidimensional vector [0.1, 0.3, 0.7, 0.2, 0.1], which facilitates the model's access to a unified format input. This structure has a fixed order and normalization processing, which can support subsequent model inference calls.

[0029] S30: Input the multidimensional context vector into the pre-trained dynamic risk assessment model to calculate the trust score of the access request.

[0030] Specifically, by remotely invoking a dynamic risk assessment model deployed in the cloud, a random forest structure is used to classify and judge the leaf nodes of the input multi-dimensional scenario vector one by one through multiple decision tree paths and execute a voting mechanism. The voting result is then mapped to a continuous trust score between 0 and 100, representing the credibility of the current access request. For example, the input vector [0.9, 0.8, 0.3, 0.7, 0.6] outputs a score of 85 after model processing, indicating high credibility. The model structure has been fully learned during the training phase using historical data with authorization success and failure labels, and has the ability to accurately distinguish between low-credibility access and reasonable access behavior. The scoring output process is stable and traceable. By remotely invoking the dynamic risk assessment model deployed on the cloud server, this model receives multi-dimensional scenario vector input from the door lock and performs all assessment processes, including feature normalization, scoring factor matching, inference path decision-making, and trust score output, uniformly in the cloud. This avoids placing complex model calculations locally on the door lock, thereby ensuring computational efficiency and model security, while improving the resource utilization of the terminal device.

[0031] S40: Automatically execute adaptive authorization decisions based on the trust score range.

[0032] Specifically, the system determines access based on a trust score and a preset threshold, with a high trust range of [80–100], a medium trust range of [50–80], and a low trust range of [0–50]. If the current request trust score is 90, the system authorizes unlocking directly without user intervention. If the trust score is 65, the action logic triggers a semantic verification process, using voice dialogue to determine the validity of the access request before deciding whether to unlock. If the score is only 30, access is directly denied, and an "invalid request" message is displayed on the screen. The access event is also marked as a potential anomaly, and a real-time alert notification is pushed to the user via the app, such as "Stranger accessing at night, user is not currently home." This tiered authorization strategy can improve user experience flexibility while ensuring security.

[0033] S50: Records the context information, calculation process, interaction content and authorization results of the entire access request process, and generates a traceable authorization log with a timeline structure.

[0034] Specifically, all key data is logged throughout the entire access processing cycle, including access trigger time, original fields of multidimensional contextual data, extracted content of scoring factors, details of multidimensional vector composition, model calculation path and trust value output results, as well as recognized text of semantic interaction with visitors, semantic matching scores, and the final unlock / deny command results. All data is integrated into structured records in chronological order and bound to a unique access event number, which is then written to cloud log storage. For example, when a user reviews an authorization behavior at a certain time in the early morning of a certain day, they can access the access record to view the entire process evidence, such as facial recognition confidence, GPS location, semantic verification score, and model judgment score. This structure ensures that the entire event process is reproducible, traceable, and verifiable.

[0035] In one embodiment, such as Figure 2 As shown, in step S10, when an access request without preset permissions is detected, a multi-dimensional context information collection instruction is triggered to obtain multi-dimensional context information corresponding to the current access request, forming multi-dimensional context data, specifically including: S11: Real-time acquisition of access scene images through door lock camera, and analysis of access scene images based on image recognition algorithm to extract visitor feature data to generate environmental context information. Among them, visitor feature data includes visitor facial features, head orientation, occlusion behavior and clothing type.

[0036] Specifically, while capturing images of the visitor scene in real time through the door lock camera, the system calls an image recognition algorithm to process each image frame by frame, extracting visitor feature data including facial feature points, head direction angle, occlusion detection, and clothing category recognition. This data is then labeled as structured image context tags. For example, when the system identifies a visitor who is obscuring their face (such as wearing a mask, lowering their head, or using a mobile phone to cover their face) or wearing unfamiliar uniforms (such as a deliveryman's jacket or a food delivery platform uniform), the system will automatically mark the behavior as "suspicious," thereby improving the accuracy and interpretability of environmental context information.

[0037] S12: Collect ambient sound and visitor voice information at the site via microphone, and perform keyword recognition and audio status analysis on the visitor voice information to supplement environmental context information.

[0038] Specifically, the on-site audio signals continuously collected by the microphone are first filtered out by a noise reduction module to remove background noise. Then, the speech segment is separated from the non-speech segment. Keyword extraction and emotion and tone recognition analysis are performed on the speech segment. For example, if "I am Li Ming and I have made an appointment with Mr. Zhang" is identified, it can be marked as a semantically clear reason for the visit. At the same time, audio state indicators such as speech rate, volume, and stress distribution are extracted to determine whether there are abnormal emotional behaviors such as tension or hesitation. Then, the background sounds in the non-speech segment (such as vehicle sounds and conversations among multiple people) are combined to construct an environmental sound label to supplement the image analysis results.

[0039] S13: Obtain the current access timestamp, break down and analyze the access timestamp, and construct time context information in the form of time tags.

[0040] Specifically, the current access time is obtained by accessing the device's system timestamp, and this time is broken down into structured time fields such as year, month, day, weekday, hour, and minute. These fields are then matched with built-in holidays and user-personalized schedule rules to generate time tags. For example, access behavior initiated at 2:00 AM on Saturday will be marked as "late-night access" and "non-working hours," thereby improving the ability to identify abnormal behavior in time contexts and enhancing the sensitivity of subsequent risk assessment models to time-related risks.

[0041] S14: Read the user's calendar events and communication keywords through a secure communication interface with the user's mobile terminal to generate user data context information.

[0042] Specifically, the system accesses the user's calendar and SMS application data through the established encrypted communication interface with the user's mobile terminal, reads the schedule and communication keyword summaries of the most recent 72 hours, and calls the keyword matching algorithm in the local cache to analyze whether there is a semantic relationship with the current visit. For example, if the system identifies that the calendar event title contains "water pipe repair" or the communication record contains "Master Wang came to check the water meter at noon", the system will generate a "predicted visitor matching" user data context label and assign it a high credibility reference weight.

[0043] S15: Obtain GPS data from the user's mobile terminal and determine the user's geographical location status based on the GPS data, thereby forming location context information.

[0044] Specifically, by establishing a location data access connection with the user's mobile terminal's location service module, the system obtains the current GPS location information and calculates the geospatial distance with the door lock's location coordinates. When the distance is less than a preset threshold and the movement speed is lower than a set value, it is marked as "the homeowner is at home." Conversely, if the distance is more than 5 kilometers or the location is in a different workplace area such as an "office building" or "airport," it is marked as "the homeowner is away." For example, if the homeowner's mobile phone location is detected as another city, the current door lock access will be defaulted to the "homeowner is not present" state by the system, enhancing the judgment of location context information.

[0045] S16: After statistically integrating environmental context information, time context information, user data context information, and location context information, a multi-dimensional context vector is formed.

[0046] Specifically, the environmental context information, time context information, user data context information, and location context information generated above will first enter a structured preprocessing flow. Historical data interpolation will be performed on missing fields, moving average correction will be applied to deviation values, and Min-Max normalization will be uniformly performed. Subsequently, a one-dimensional vector encoder will be used to convert the text, image, and audio tag data into dense feature vectors, and then concatenate them into a multi-dimensional context vector of a unified dimension. This vector structure is, for example, [0.78, 0.35, 0.65, 0.1, 0.93, ...], where each item represents sub-dimensional indicators such as time matching degree, semantic keyword similarity, and location credibility level, providing a standardized input format for subsequent model processing.

[0047] In one embodiment, such as Figure 3 As shown, in step S20, the multidimensional context data is preprocessed to obtain a multidimensional context vector, specifically including: S21: Perform feature extraction on the multidimensional contextual data to extract multiple scoring factors to characterize the credibility of the visit. The scoring factors include: Visitor identity tags are generated by comparing visitor facial features with a user-preset trusted facial image database.

[0048] Specifically, by performing a feature matching algorithm on the facial feature vectors in the multidimensional contextual data, the facial feature encoding of the current visitor is compared one by one with the locally cached trusted facial image library, and the cosine similarity is calculated to quantify the similarity level. The similarity result is then mapped to an identity label value. For example, when the similarity is higher than a preset threshold, the identity label is recorded as "highly trustworthy visitor", while when the similarity is low, it is marked as "unknown visitor" or "low trustworthy visitor". In actual operation, if the features of a family member are found to be highly consistent with the image library record, the overall trustworthiness can be quickly improved, thereby improving the accuracy of identifying familiar visitors.

[0049] The time matching status is generated by comparing the current access time with the time period of the user's scheduled calendar event.

[0050] Specifically, by parsing the current access timestamp and comparing it with appointment events in the user's calendar within the next 48 hours, the system calculates whether the access time falls within a certain appointment service time window and generates a time matching score. For example, if the current time is within the range of "housekeeping appointment 14:00–15:00", the access will be marked as "high time matching". Conversely, if the access occurs during a sensitive period set by the user in the long term, such as 0:00 to 5:00 in the morning, it will be automatically marked as "time abnormal", thus realizing the dynamic quantification of the rationality of the access time.

[0051] The location proximity marker is generated based on the distance between the user's GPS data and the door lock location.

[0052] Specifically, by reading the real-time GPS information of the user's mobile terminal and performing geographical distance calculation, the calculation result is compared with the straight-line distance of the fixed location coordinates of the door lock and a location distance marker is generated. When the distance is less than 50 meters and the moving speed is lower than a set threshold, it can be marked as "the user is near home". If the user's location is in an office area or business district several kilometers away, it is marked as "the user is far away". For example, when the user's location is detected to be in another city, the credibility of the current access will be reduced accordingly, thereby improving the ability of location factors to judge the rationality of the access.

[0053] The semantic matching score is generated by calculating the similarity between visitor voice information and user communication keywords.

[0054] Specifically, the system performs speech recognition on the visitor's speech content to obtain text results and extract core phrases. Then, it calculates word vector similarity with the user's communication keywords within the past week. A higher semantic matching score will be generated when the content of the visitor's statement is highly related to the semantics of the user's recent communication. For example, if the visitor's speech content contains words such as "plumbing repair" or "appointment" and similar keywords appear in the user's recent communication records with the repair company, the similarity will be improved, thereby enhancing the judgment on the authenticity of the visit reason.

[0055] The behavioral pattern matching score is generated based on the similarity analysis between the current access context and historical successful authorization behaviors.

[0056] Specifically, by calling the successful authorization records of the past three months in the historical access database and generating behavioral pattern vectors, the system performs similarity analysis between the multidimensional features of the current visit and these behavioral patterns. When the visitor's behavioral features, visitor's appearance time, user location status, and environmental indicators are all highly consistent with the historical normal visitor behavior, the system will output a high behavioral pattern matching score. For example, if a frequently visiting relative usually visits on weekend afternoons and speaks in a stable manner, then their visit features will show good consistency with historical data, thus improving credibility.

[0057] S22: The scoring factors are structured and encoded according to the preset feature dimensions to form a multi-dimensional context vector for model input.

[0058] Specifically, by performing standardization and feature dimension encoding on the aforementioned scoring factors such as identity tags, time matching status, location proximity tags, semantic matching scores, and behavioral pattern matching scores, they are converted into a dense vector structure with a unified dimension. During the encoding process, fixed index positions are assigned to different features to form a stable input format, so that the final multidimensional context vector can be input to the dynamic risk assessment model deployed in the cloud in a manner such as [0.82, 0.31, 0.67, 0.15, 0.93, ...], ensuring that the model can achieve real-time credibility assessment based on structured vectors.

[0059] In one embodiment, such as Figure 4 As shown, this context-aware smart lock dynamic authorization control method also includes: S301: Construct a training dataset based on historical access request records. The training dataset includes a multi-dimensional context vector and access result labels formed by extracting and encoding the multi-dimensional context information corresponding to each access request. The multi-dimensional context information includes time context, environmental context, user data context and location context. The access result labels include authorization success, authorization failure, user denial, no response and abnormal alarm.

[0060] Specifically, a training dataset is constructed based on historical access request records. Data query operations are performed to obtain historical data including access time, visitor behavior, user terminal status, and access location. Data samples for training are extracted from this data. During this process, the data is labeled and divided according to timeline and behavior type. Access result labels include status information such as authorization success, authorization failure, user active rejection, no response, or alarm trigger. Each sample data corresponds to a multi-dimensional context vector and a clear result label. The sample data is uploaded to the cloud for cleaning and structuring to form a training dataset for subsequent model training.

[0061] S302: The initial classification model is constructed using the random forest algorithm, and the training dataset is used as input to perform a supervised learning process to optimize the model parameters, resulting in a pre-trained dynamic risk assessment model.

[0062] Specifically, a random forest algorithm is used to build an initial classification model. Multiple rounds of supervised learning are performed to improve the model's recognition ability. In the initial stage, basic parameters of the random forest, such as the number of trees, maximum depth, and sample splitting criteria, are set. The model is trained using a training dataset. Cross-validation is performed during training to prevent overfitting, and model parameters are continuously adjusted to improve prediction accuracy. Finally, a dynamic risk assessment model with an acceptable input of multi-dimensional context vectors and an output of trust scores is generated. The training process and model validation are both completed in the cloud to ensure unified management of training resources and model versions.

[0063] S303: Deploy the pre-trained dynamic risk assessment model to the cloud server. When it receives the real-time generated multi-dimensional context vector, it outputs a continuous trust score between 0 and 100. The trust score is used to drive the execution logic of subsequent adaptive authorization decisions.

[0064] Specifically, the trained dynamic risk assessment model is uploaded and deployed to a cloud server. When an access request is triggered, a multi-dimensional context vector generated by an edge terminal device is received in real time. The model inference operation is performed to output a trust score, which is a continuous value between 0 and 100, used to characterize the credibility of the current access request. This process realizes end-cloud collaboration. The model inference logic is completed in the cloud, and the inference result is sent to the edge terminal device through a secure channel to guide the execution of subsequent authorization policies, ensuring that the dynamic and intelligent nature of access control is achieved while ensuring data processing efficiency.

[0065] In one embodiment, such as Figure 5 As shown, in step S40, the adaptive authorization decision is automatically executed based on the trust score range, specifically including: S41: When the trust score is in the preset high trust range, execute the automatic authorization operation, directly send the unlocking command to the door lock control module, and simultaneously record the authorization pass event.

[0066] Specifically, after receiving the trust score output from the dynamic risk assessment model, a credibility interval judgment operation is performed. When the judgment result indicates that the current trust score is within the preset high credibility interval, an automatic authorization process is immediately triggered, generating and issuing structured control instructions to the door lock control module to complete the door lock unlocking action. At the same time, the key data fields of this authorization event, including trust score, trigger time, visitor identification, and operation feedback, are encapsulated into a structured event log and written to a local or cloud log storage node to achieve closed-loop tracking of the event process and subsequent audit support. For example, when a visitor's trust score reaches 92 points, the authorization process can complete the instruction issuance and log writing operations within 500ms.

[0067] S42: When the trust score is in the preset medium trust range, trigger the semantic verification assisted authorization process.

[0068] Specifically, when the trust score is in the medium-credibility range, the auxiliary verification branch logic is executed, which calls the voice guidance module to play a preset semantic prompt or text prompt to the visitor to request further verification of identity information. At the same time, the voice data acquisition channel is initialized and the response data stream is waited for. When the visitor responds by voice, the voice recognition and semantic understanding process is executed to extract keyword data from the voice content and generate a response data package for subsequent similarity judgment operations. This process aims to guide the user to provide semantic information in partially credible but not completely uncertain situations to enhance the contextual basis of the model's decision-making. For example, when the trust score is 67, the visitor is prompted to explain the reason for the visit and state the name or commonly used phrases related to the homeowner.

[0069] S43: When the trust score is in the preset low trust range, reject the current access request, output a rejection prompt to the visitor, and push a risk notification to the user's mobile terminal to remind them of potential abnormal access behavior.

[0070] Specifically, when the trust score is in the low confidence range, the security denial process is immediately initiated, generating a denial prompt to control the door lock interface to play or display denial voice and image content. At the same time, the alert module is invoked to push a high-risk access alert to the user's mobile terminal, including information such as trigger time, risk level, visitor image screenshot and trust score. The alert information is transmitted through an encrypted communication channel and supports quick user feedback and processing. In addition, after the denial operation is completed, the abnormal behavior is synchronously recorded in the local behavior log and marked as a high-priority record for subsequent policy adjustments. For example, if an unfamiliar face is detected repeatedly trying to unlock the door late at night and the score is below 40, the entire process of locking, prompting and pushing can be automatically completed.

[0071] In one embodiment, such as Figure 6 As shown, in step S42, the semantic verification-assisted authorization process is triggered, which specifically includes: S421: When the trust score is in the preset medium trust range, initiate a verification prompt voice to the visitor, prompting the visitor to state the reason for the visit or verify the password.

[0072] Specifically, after detecting that the trust score falls into the middle confidence range, a semantic verification initialization operation is performed. A preset verification prompt voice is played through the voice output interface to guide the visitor to actively state the reason for the visit or to give a pre-agreed verification password. During the playback of the prompt voice, a listening channel for the visitor's voice input is established and the input buffer is kept in a writable state to ensure that the visitor's natural language response content can be fully received. For example, when a visitor's trust score is 62 points in the evening, a prompt voice such as "Please state the purpose of your visit or give your verification password" will be played, thereby prompting the visitor to provide additional semantic information for subsequent verification analysis.

[0073] S422: Collect visitor response speech and perform speech recognition and natural language semantic analysis to extract keyword content from the response speech.

[0074] Specifically, after receiving the visitor's response voice data stream, a voice parsing process is executed. The continuous voice signal is transcribed into a text sequence using an acoustic model. Then, based on semantic analysis algorithms, the text sequence is syntactically split, lexical filtering, and semantic extraction. Keyword content that can represent the semantic core is extracted from the visitor's statement. These keywords may include semantic fragments such as "looking for someone," "express delivery," "takeout," and "repair." In this way, standardized semantic units are constructed for subsequent similarity comparison. For example, when a visitor responds "I'm here to repair the water pipes," keywords such as "repair" and "water pipes" can be extracted after parsing for subsequent comparison.

[0075] S423: Perform semantic similarity matching between the keyword content and the communication keywords, calendar event descriptions, or trust phrases set by the authorizer extracted from the user's mobile terminal to obtain a semantic similarity matching score.

[0076] Specifically, after obtaining the keyword content from the visitor, a semantic similarity calculation process is executed. A local or cloud-based semantic vector matching model is called to perform bidirectional vector encoding on the visitor's keywords and communication keywords, calendar event descriptions, or commonly used semantic phrases extracted from the user's historical behavior. The semantic similarity score between the two is then calculated. The similarity calculation can use methods such as cosine similarity, semantic embedding distance, or deep matching algorithms to measure the closeness between the visitor's semantics and the user's expected context. For example, when a "water pipe repair" event appears in the user's calendar record, the semantic similarity of the visitor's statement "water pipe repair" may increase to a higher score, thereby supporting the judgment of auxiliary verification.

[0077] S424: When the semantic similarity matching score reaches the preset verification threshold, execute the auxiliary authorization instruction to control the door lock to unlock; otherwise, keep the door lock closed and send a notification to the user to request confirmation.

[0078] Specifically, after obtaining the semantic similarity matching score, a threshold judgment logic is executed, comparing the matching score with a preset verification threshold. When the score is higher than the verification threshold, an auxiliary authorization command is generated and the door lock is unlocked through the control interface. At the same time, the successful verification event is recorded for subsequent log tracking. When the score is lower than the verification threshold, the door lock remains closed and a notification is immediately sent to the user's mobile terminal, requesting the user to manually confirm the access. This ensures both security and access convenience. For example, when the similarity score is only 0.35 and the threshold is set to 0.6, a reminder message such as "Someone attempted to access but the semantic verification failed. Please confirm whether to authorize" will be sent directly to the user.

[0079] In one embodiment, such as Figure 7 As shown, this context-aware smart lock dynamic authorization control method also includes: S60: When abnormal access behavior is detected, the local sound and light alarm device is automatically triggered and a risk warning message is sent to the user's mobile terminal. At the same time, the current access record is marked as a high-risk event and included in the authorization log. Abnormal access behavior includes consecutive failed authorization attempts by the same identity within a short period of time, access time within a sensitive time period set by the user, and face occlusion or multiple people gathering in the image recognition result.

[0080] Specifically, after determining that the current access is an abnormal access behavior, the abnormal event handling process is executed. The real-time analysis results of the access data trigger the activation command of the local audible and visual alarm device, causing the alarm device to emit a high-brightness flash or a high-decibel alarm sound in a very short time to create a deterrent effect on site. At the same time, a structured risk warning message is generated for the user's mobile terminal and pushed in real time through an encrypted communication link, so that the user can know about the abnormal access situation at the first time. On this basis, the key fields of the current access record are marked with risks and written into the high-risk event area of ​​the authorization log along with the event time, judgment basis and access behavior fragment, so as to facilitate subsequent tracking, evidence collection and model correction. For example, when an identity is detected to have entered the wrong access verification information three times in a row in the early morning hours, accompanied by facial occlusion in the image, an audible and visual alarm will be immediately executed and a "detected abnormal access attempt" notification will be pushed to the user. At the same time, the access event will be recorded as a high-risk access for subsequent audit analysis.

[0081] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0082] In one embodiment, a context-aware intelligent door lock dynamic authorization control system is provided, which corresponds one-to-one with the context-aware intelligent door lock dynamic authorization control method described in the above embodiments. For example... Figure 8 As shown, this context-aware intelligent door lock dynamic authorization control system includes an access triggering module, a data preprocessing module, a risk scoring module, an authorization decision module, and a log recording module. Detailed descriptions of each functional module are as follows: The access triggering module is used to trigger a multi-dimensional context information collection instruction when an access request without preset permissions is detected, to obtain multi-dimensional context information corresponding to the current access request, and to form multi-dimensional context data. The multi-dimensional context information includes time context, environmental context, user data context, and location context. The data preprocessing module is used to preprocess multidimensional contextual data to obtain multidimensional contextual vectors; The risk scoring module is used to input multi-dimensional context vectors into a pre-trained dynamic risk assessment model to calculate the trust score of the access request. The authorization decision module is used to automatically execute adaptive authorization decisions based on the trust score range. The logging module is used to record contextual information, calculation process, interaction content and authorization results of the entire access request process, and generate a traceable authorization log with a timeline structure.

[0083] Optionally, the access triggering module includes: The image acquisition and analysis submodule is used to acquire images of the access scene in real time through the door lock camera, and analyze the access scene images based on the image recognition algorithm to extract visitor feature data in order to generate environmental context information. The visitor feature data includes visitor facial features, head orientation, occlusion behavior and clothing type. The audio analysis submodule is used to collect ambient sound and visitor voice information at the site via microphone, and to perform keyword recognition and audio state analysis on the visitor voice information to supplement environmental context information; The timestamp submodule is used to obtain the current access timestamp, break down and analyze the access timestamp, and construct time context information in the form of timestamps; The calendar communication interface submodule is used to read the user's calendar events and communication keywords through a secure communication interface with the user's mobile terminal in order to generate user data context information; The location determination submodule is used to acquire GPS data from the user's mobile terminal and determine the user's geographical location status based on the GPS data, thereby forming location context information; The context integration submodule is used to statistically integrate environmental context information, time context information, user data context information, and location context information to form a multi-dimensional context vector.

[0084] Optionally, the data preprocessing module includes: The feature extraction submodule performs feature extraction operations on multidimensional contextual data, extracting multiple scoring factors to characterize access credibility. These scoring factors include: Visitor identity tags are generated by comparing visitor facial features with a user-preset trusted facial image database. The time matching status is generated by comparing the current access time with the time period of the user's scheduled calendar event. The location proximity marker is generated based on the distance between the user's GPS data and the door lock location. The semantic matching score is generated based on the similarity calculation between visitor voice information and user communication keywords; Behavioral pattern matching scores are generated based on similarity analysis between the current access context and historical successful authorization behaviors. The vector encoding submodule is used to structure and encode the rating factors according to the preset feature dimensions to form a multi-dimensional context vector for model input.

[0085] Optionally, this context-aware intelligent door lock dynamic authorization control system also includes: The training set construction module is used to build a training dataset based on historical access request records. The training dataset includes a multi-dimensional context vector extracted and encoded from the multi-dimensional context information corresponding to each access request, and an access result label. The multi-dimensional context information includes time context, environmental context, user data context, and location context. The access result label includes authorization success, authorization failure, user denial, no response, and abnormal alarm. The model training module is used to build an initial classification model using the random forest algorithm, and with the training dataset as input, to perform a supervised learning process to optimize the model parameters and obtain a pre-trained dynamic risk assessment model. The model deployment and scoring module is used to deploy the pre-trained dynamic risk assessment model to the cloud server. When it receives the real-time generated multi-dimensional context vector, it outputs a continuous trust score between 0 and 100. The trust score is used to drive the execution logic of subsequent adaptive authorization decisions.

[0086] Optionally, the authorization decision module includes: The high-trust authorization submodule is used to perform automatic authorization when the trust score is in the preset high-trust range. It directly sends the unlocking command to the door lock control module and records the authorization pass event simultaneously. The medium-trust verification submodule is used to trigger the semantic verification-assisted authorization process when the trust score is within the preset medium-trust range; The low-trust rejection submodule is used to reject the current access request when the trust score is in the preset low-trust range, output a rejection prompt to the visitor, and push a risk notification to the user's mobile terminal to alert them to potential abnormal access behavior.

[0087] Optionally, the trusted verification submodule includes: The voice prompt generation unit is used to initiate a verification prompt voice to the visitor when the trust score is in the preset medium trust range, prompting the visitor to state the reason for the visit or verify the password; The speech recognition and analysis unit is used to collect visitor response speech and perform speech recognition and natural language semantic analysis to extract keyword content from the response speech. The semantic matching scoring unit is used to perform semantic similarity matching between keyword content and communication keywords, calendar event descriptions, or trust phrases set by the authorizer extracted from the user's mobile terminal, and obtain a semantic similarity matching score. The auxiliary authorization judgment unit is used to execute the auxiliary authorization instruction to control the door lock to unlock when the semantic similarity matching score reaches the preset verification threshold; otherwise, it keeps the door lock closed and sends a notification to the user to request confirmation.

[0088] Optionally, this context-aware intelligent door lock dynamic authorization control system also includes: The abnormal behavior response module is used to automatically trigger the local audible and visual alarm device and send risk warning information to the user's mobile terminal when abnormal access behavior is detected. At the same time, the current access record is marked as a high-risk event and included in the authorization log. Abnormal access behavior includes consecutive failed authorization attempts by the same identity within a short period of time, access time within a sensitive time period set by the user, and facial occlusion or multiple people gathering in the image recognition result.

[0089] For specific limitations regarding the context-aware intelligent door lock dynamic authorization control system, please refer to the limitations of the context-aware intelligent door lock dynamic authorization control method described above, which will not be repeated here. Each module in the aforementioned context-aware intelligent door lock dynamic authorization control system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the operations corresponding to each module.

[0090] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system can be divided into different functional units or modules to complete all or part of the functions described above.

[0091] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A context-aware dynamic authorization control method for smart door locks, characterized in that, The aforementioned context-aware smart lock dynamic authorization control method includes: When an access request without preset permissions is detected, a multi-dimensional context information collection instruction is triggered to obtain multi-dimensional context information corresponding to the current access request, forming multi-dimensional context data. The multi-dimensional context information includes time context, environmental context, user data context, and location context. The multidimensional context data is preprocessed to obtain a multidimensional context vector; The multidimensional context vector is input into a pre-trained dynamic risk assessment model to calculate the trust score of the access request; Automatically execute adaptive authorization decisions based on the trust score range; Record the context information, calculation process, interaction content and authorization results of the entire access request process, and generate a traceable authorization log with a timeline structure.

2. The context-aware intelligent door lock dynamic authorization control method according to claim 1, characterized in that, When an access request without preset permissions is detected, a multi-dimensional context information collection instruction is triggered to obtain multi-dimensional context information corresponding to the current access request, forming multi-dimensional context data, specifically including: The system uses a door lock camera to capture images of the visit scene in real time, and analyzes these images based on an image recognition algorithm to extract visitor feature data in order to generate environmental context information. The visitor feature data includes visitor facial features, head orientation, occlusion behavior, and clothing type. The system collects ambient sound and visitor voice information at the site using a microphone, and performs keyword recognition and audio state analysis on the visitor voice information to supplement the environmental context information. Obtain the current access timestamp, decompose and analyze the access timestamp, and construct time context information in the form of time tags; By reading the user's calendar events and communication keywords through a secure communication interface with the user's mobile terminal, user data context information is generated; The system acquires GPS data from the user's mobile terminal and determines the user's geographical location status based on the GPS data, thereby forming location context information. The environmental context information, time context information, user data context information, and location context information are statistically integrated to form a multidimensional context vector.

3. The context-aware intelligent door lock dynamic authorization control method according to claim 1, characterized in that, The step of preprocessing the multidimensional context data to obtain a multidimensional context vector specifically includes: Feature extraction is performed on the multidimensional contextual data to extract multiple scoring factors used to characterize access credibility. These scoring factors include: Visitor identity tags are generated by comparing visitor facial features with a user-preset trusted facial image database. The time matching status is generated by comparing the current access time with the time period of the user's scheduled calendar event. The location proximity marker is generated based on the distance between the user's GPS data and the door lock location. The semantic matching score is generated based on the similarity calculation between visitor voice information and user communication keywords; Behavioral pattern matching scores are generated based on similarity analysis between the current access context and historical successful authorization behaviors. The scoring factors are structured and encoded according to preset feature dimensions to form a multidimensional context vector for model input.

4. The context-aware intelligent door lock dynamic authorization control method according to claim 1, characterized in that, The aforementioned context-aware smart lock dynamic authorization control method further includes: A training dataset is constructed based on historical access request records. The training dataset includes a multidimensional context vector and access result labels formed by extracting and encoding multidimensional context information corresponding to each access request. The multidimensional context information includes time context, environmental context, user data context and location context. The access result labels include authorization success, authorization failure, user rejection, no response and abnormal alarm. An initial classification model is constructed using the random forest algorithm, and a supervised learning process is performed using the training dataset as input to optimize the model parameters, thereby obtaining the pre-trained dynamic risk assessment model. The pre-trained dynamic risk assessment model is deployed to a cloud server. When a multi-dimensional context vector generated in real time is received, a continuous trust score between 0 and 100 is output. The trust score is used to drive the execution logic of subsequent adaptive authorization decisions.

5. The context-aware intelligent door lock dynamic authorization control method according to claim 1, characterized in that, The automatic execution of adaptive authorization decisions based on the trust score range specifically includes: When the trust score is in the preset high confidence range, an automatic authorization operation is performed, directly sending an unlocking command to the door lock control module and simultaneously recording the authorization pass event; When the trust score is within the preset medium trust range, the semantic verification-assisted authorization process is triggered; When the trust score is in a preset low trust range, the current access request is rejected, a rejection prompt is displayed to the visitor, and a risk notification is pushed to the user's mobile terminal to alert them to potential abnormal access behavior.

6. The context-aware intelligent door lock dynamic authorization control method according to claim 5, characterized in that, The triggering semantic verification-assisted authorization process specifically includes: When the trust score is in the preset medium trust range, a verification prompt voice is issued to the visitor, prompting the visitor to state the reason for the visit or to verify the password; Collect visitor response speech and perform speech recognition and natural language semantic analysis to extract keyword content from the response speech; The semantic similarity of the keyword content is matched with the communication keywords, calendar event descriptions or trust phrases set by the authorizing person extracted from the user's mobile terminal to obtain a semantic similarity matching score. When the semantic similarity matching score reaches the preset verification threshold, an auxiliary authorization instruction is executed to control the door lock to unlock; otherwise, the door lock remains closed and a notification is sent to the user to request confirmation.

7. The context-aware intelligent door lock dynamic authorization control method according to claim 1, characterized in that, The aforementioned context-aware smart lock dynamic authorization control method further includes: When abnormal access behavior is detected, the local sound and light alarm device is automatically triggered and a risk warning message is sent to the user's mobile terminal. At the same time, the current access record is marked as a high-risk event and included in the authorization log. The abnormal access behavior includes consecutive failed authorization attempts by the same identity within a short period of time, access time within a sensitive time period set by the user, and face occlusion or multiple people gathering in the image recognition result.

8. A context-aware intelligent door lock dynamic authorization control system, characterized in that, The context-aware intelligent door lock dynamic authorization control system includes: The access triggering module is used to trigger a multi-dimensional context information collection instruction when an access request without preset permissions is detected, to obtain multi-dimensional context information corresponding to the current access request, and to form multi-dimensional context data. The multi-dimensional context information includes time context, environmental context, user data context, and location context. The data preprocessing module is used to preprocess the multidimensional context data to obtain a multidimensional context vector; The risk scoring module is used to input the multidimensional context vector into a pre-trained dynamic risk assessment model to calculate the trust score of the access request. The authorization decision module is used to automatically execute adaptive authorization decisions based on the range to which the trust score belongs; The logging module is used to record contextual information, calculation process, interaction content and authorization results of the entire access request process, and generate a traceable authorization log with a timeline structure.

Images