Unlocking mode switching method and device of intelligent lock

Abstract

The invention provides an unlocking mode switching method and device of an intelligent lock, and relates to the technical field of emergency unlocking, and the key points of the technical scheme are as follows: monitoring a network connection state, and receiving a state signal; switching to an emergency mode when network connection interruption exceeds a preset duration threshold value or a system fault signal is received; when the low-precision face recognition comparison locally executed by the intelligent lock succeeds, an unlocking instruction is executed, and after the intelligent lock is unlocked, emergency unlocking event information is locally stored in the intelligent lock; and after the central management system recovers normal operation, the emergency unlocking event information stored locally is uploaded to the central management system, and a survey instruction is generated. The unlocking mode switching method and device for the intelligent lock, provided by the invention, have the advantage of ensuring that a resident can quickly and effectively unlock a door lock in an extreme scene that a central management system breaks down.

Description

Technical Field

[0001] The present application relates to the field of emergency unlocking technology, and more specifically, to a method and device for switching unlocking modes of a smart lock. Background Art

[0002] At present, high-end apartment buildings generally use face recognition smart locks to improve the security level and intelligent experience of the living environment. This type of smart lock is usually integrated into the central management system to build a unified intelligent security system. The central management system not only bears the heavy responsibility of security monitoring, but also is responsible for the access control and identity authentication of residents. Through the network, the central management system and each household's smart door lock are linked in real time to achieve the purpose of centralized control and management. In daily use scenarios, residents use the face recognition module on the smart lock to authenticate their identities. The smart lock will upload the collected face information to the central management system, and the system will perform high-precision face comparison and identity verification. Once the verification process is passed, the central management system sends an unlocking command to the smart lock, and the door lock is opened immediately, thereby ensuring the intelligence and safety of the door lock opening operation.

[0003] However, the central management system that high-end apartment buildings rely on is not absolutely stable and reliable. In actual operation, it may encounter various types of local failures. For example, the network connection in a specific area is interrupted, the server runs slowly due to excessive load, or the software system has an abnormality. When a local failure occurs in the central management system, the response speed of the building access control system will be significantly reduced, or even completely lost, which directly causes those face recognition smart locks that rely on the central management system for network identity authentication to fail to work properly. What's more tricky is that in order to prevent security vulnerabilities that may occur during system failures, some security systems will adopt more stringent security strategies when a failure is detected. For example, the local backup unlocking method of the smart lock, such as password unlocking, fingerprint unlocking, etc., is forced to lock to minimize the risk of potential illegal intrusion.

[0004] In the above scenario, the residents' facial recognition smart locks lose their original unlocking function because they cannot establish a connection with the central management system to complete identity authentication. At the same time, the local backup unlocking method is forcibly locked by the system, which makes the residents completely lose the effective way to open their own door locks without relying on the central management system to verify the network and without unlocking the system. This not only brings great inconvenience to the residents' daily entry and exit. Therefore, how to design a reliable emergency unlocking solution that can not only ensure the safety of residents at home but also meet the system security management requirements in the scenario of a partial failure of the central management system to achieve fast and effective door lock opening has become a key issue that needs to be solved in the current field of smart door lock technology.

[0005] In view of the above problems, there is an urgent need for improvement in the existing technology. Summary of the Invention

[0006] The purpose of this application is to provide a method and device for switching the unlocking mode of an intelligent lock, which can ensure that residents can quickly and effectively unlock the door lock in extreme scenarios where the central management system fails.

[0007] In the first aspect, this application provides a method for switching the unlocking mode of an intelligent lock, and the technical solution is as follows: The method includes: Monitoring the network connection status between the intelligent lock and the central management system, and receiving the status signal from the central management system; When the network connection is interrupted for more than a preset duration threshold, or a system failure signal sent by the central management system is received, switch to the emergency mode to deactivate the high-precision face recognition originally executed in the central management system and enable the low-precision face recognition executed locally on the intelligent lock; When the low-precision face recognition executed locally on the intelligent lock is successfully compared, execute the unlocking instruction, and after the intelligent lock is opened, store the emergency unlocking event information locally on the intelligent lock; When the central management system resumes normal operation, upload the locally stored emergency unlocking event information to the central management system and generate an investigation instruction.

[0008] Further, in this application, the step of switching to the emergency mode to deactivate the high-precision face recognition originally executed in the central management system and enable the low-precision face recognition executed locally on the intelligent lock includes: Switch to the emergency mode, and control the communication unit to switch the recognized face image information from originally being sent to the central management information to being sent to the local processing unit; Control the local processing unit of the intelligent lock to load a low-precision algorithm to perform face recognition on the obtained face image information.

[0009] Further, in this application, the low-precision algorithm loaded by the local processing unit of the intelligent lock is a personalized algorithm optimized through parameter iteration.

[0010] Further, in this application, it further includes: After the intelligent lock executes the unlocking instruction through the high-precision face recognition executed by the central management system, extract the face feature information recognized by the high-precision face recognition; Perform parameter iteration optimization on the low-precision algorithm according to the face feature information and the recognition passed result to obtain a personalized algorithm.

[0011] Further, in this application, the personalized algorithm is implemented based on a neural network.

[0012] Further, in the present application, the personalized algorithm is implemented based on the update of the face template. The process of iteratively optimizing the parameters of the low-precision algorithm according to the face feature information and the recognized result to obtain the personalized algorithm includes: Obtain the face feature information of multiple successful recognitions, and obtain the corresponding face feature vectors for the face feature information of multiple successful recognitions; Calculate the variances of each dimension of the face feature vectors for multiple times, and determine the dimensions with variances lower than the preset threshold as stable features; Perform weighted averaging on the face feature vectors in chronological order, where the latest features obtain higher weights, to form a time-weighted feature vector; Combine the recognized stable features with the time-weighted feature vector to update the original local face template; Based on the importance of the stable features, adjust the weight coefficients of each feature dimension in the low-precision algorithm, and at the same time adjust the matching threshold, so as to obtain the personalized algorithm.

[0013] Further, in the present application, the steps of uploading the local stored emergency unlocking event information to the central management system and generating an investigation instruction after the central management system resumes normal operation include: Classify the emergency events; Determine the information type of the local stored emergency unlocking event information to be uploaded to the central management system according to the classification of the emergency events; Generate an investigation instruction according to the uploaded information type and the classification of the emergency events.

[0014] Further, in the present application, the steps of uploading the local stored emergency unlocking event information to the central management system and generating an investigation instruction after the central management system resumes normal operation include: When performing low-precision face recognition for emergency unlocking, record the timestamp of the unlocking operation and the recognized user identity information; Generate a query instruction to investigate the geographical location of the corresponding household user at the current timestamp; After the central management system resumes normal operation and uploads the local stored emergency unlocking event information, perform spatio-temporal consistency verification, including: Query the geographical location of the corresponding household user at the corresponding timestamp according to the query instruction; Calculate the physical distance and possible moving time between the geographical location of the corresponding household user and the residential location; Judge the spatio-temporal consistency according to the physical distance and the possible moving time. When spatio-temporal inconsistency is detected, generate a high-priority investigation instruction and mark it as a potential security risk event; Send a security alert notification to the household head.

[0015] Further, in the present application, the step of switching to the emergency mode to deactivate the high-precision face recognition originally executed in the central management system and activate the low-precision face recognition executed locally in the smart lock includes: Obtain the type of fault; According to the type of fault, adaptively switch to the emergency mode to deactivate the high-precision face recognition originally executed in the central management system and activate the low-precision face recognition executed locally in the smart lock, where the type of fault includes but is not limited to: network communication interruption, server response timeout, system operation anomaly; For the fault of network communication interruption, select an algorithm configuration with a lower computational complexity but more feature points; For the fault of server response timeout, select an algorithm configuration with a medium computational complexity and set a short recognition timeout threshold; For the fault of system operation anomaly, select an algorithm configuration with the highest security level.

[0016] In a second aspect, the present application also proposes an unlocking mode switching device for a smart lock, including: A monitoring module, configured to monitor the network connection status between the smart lock and the central management system, and receive a status signal from the central management system; A switching module, configured to switch to the emergency mode to deactivate the high-precision face recognition originally executed in the central management system and activate the low-precision face recognition executed locally in the smart lock when the network connection is interrupted for more than a preset duration threshold or a system fault signal sent by the central management system is received; An execution module, configured to execute an unlocking instruction when the low-precision face recognition executed locally in the smart lock is successfully matched, and store emergency unlocking event information locally in the smart lock after the smart lock is opened; A post-processing module, configured to upload the locally stored emergency unlocking event information to the central management system and generate an investigation instruction after the central management system resumes normal operation.

[0017] As can be seen from the above, an unlocking mode switching method and device for a smart lock provided by the present application solve the problem that the smart lock cannot be normally unlocked when the central management system fails by switching to the local low-precision face recognition mode when the network connection is interrupted or the system fails, and have the effect of ensuring that the household can quickly and effectively open the door lock in the scenario where the central management system fails or the network is interrupted. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic flowchart of an unlocking mode switching method for a smart lock provided by the present application.

[0019] Figure 2Schematic structural diagram of an unlocking mode switching device for an intelligent lock provided by this application.

[0020] Figure 3 Overall framework schematic diagram of an unlocking mode switching method for an intelligent lock provided by this application.

[0021] In the figure: 210, monitoring module; 220, switching module; 230, execution module; 240, post-processing module. Detailed implementation manners

[0022] Next, the technical solutions in this application will be clearly and completely described in conjunction with the accompanying drawings in this application. Obviously, the described embodiments are only some of the embodiments of this application, rather than all of them. Components of this application usually described and illustrated in the accompanying drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of this application that is required to be protected, but only represents selected embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative efforts fall within the scope of protection of this application.

[0023] It should be noted that: Similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, terms such as "first" and "second" are only used for differential description and cannot be understood as indicating or implying relative importance.

[0024] Referring to Figure 1 , this application proposes an unlocking mode switching method for an intelligent lock, and the method includes: S110. Monitor the network connection status between the intelligent lock and the central management system, and receive the status signal from the central management system; S120. When the network connection is interrupted for more than a preset duration threshold, or a system failure signal sent by the central management system is received, switch to the emergency mode to deactivate the high-precision face recognition originally executed in the central management system and enable the low-precision face recognition executed locally on the intelligent lock; S130. When the low-precision face recognition executed locally on the intelligent lock is successfully compared, execute the unlocking instruction, and after the intelligent lock is opened, store the emergency unlocking event information locally on the intelligent lock; S140. After the central management system resumes normal operation, upload the locally stored emergency unlocking event information to the central management system and generate an investigation instruction.

[0025] Among them, monitoring the network connection status between the monitoring intelligent lock and the central management system can be achieved by the network monitoring module in the intelligent lock periodically sending network connection detection packets to the central management system.

[0026] The reception of the status signal can be completed by the communication module of the intelligent lock listening to the information channel from the central management system in real time.

[0027] Switching to the emergency mode can be configured such that when the network monitoring module detects that no response from the central management system has been received for multiple consecutive cycles, or the communication module receives a clear fault signal sent by the central management system, the control unit of the intelligent lock issues an instruction to deactivate the original high-precision face recognition function and activate the local low-precision face recognition module.

[0028] The activation of the local low-precision face recognition module can include, for example, loading the low-precision face recognition algorithm pre-stored in the local memory and initializing the relevant hardware resources.

[0029] The local storage of the emergency unlocking event information can be achieved by creating a specific area in the local memory of the intelligent lock to record key information such as the time of the emergency unlocking and the user identity.

[0030] The generation of the investigation instruction can be set such that after the central management system resumes communication, the intelligent lock automatically uploads the local-stored emergency unlocking event information to the central management system. After receiving this information, the background server of the central management system generates a corresponding investigation task and issues the investigation instruction to the security personnel or system administrator.

[0031] Specifically, in the intelligent security system, the intelligent lock of each household maintains a network connection with the central management system. The intelligent lock continuously monitors the network status with the central management system and receives the operation status signal from the central management system.

[0032] Under normal circumstances, the intelligent lock is in the normal working mode, and the face recognition unlocking request will be sent to the central management system for high-precision comparison. When the network monitoring module of the intelligent lock detects that the network connection with the central management system is interrupted for more than a preset duration threshold of, for example, 5 minutes, or receives a system fault signal sent by the central management system, the intelligent lock immediately switches to the emergency mode.

[0033] In the emergency mode, the intelligent lock deactivates the original high-precision face recognition process that is connected to the network and instead enables the low-precision face recognition algorithm pre-installed on the local intelligent lock. When the household conducts a face recognition operation in front of the intelligent lock, the local processing unit of the intelligent lock uses the low-precision algorithm to quickly compare the captured face image. If the local low-precision face recognition comparison is successful, the intelligent lock executes the unlocking instruction and the door lock is opened. At the same time, the intelligent lock records this emergency unlocking event in the local memory, and the information can include the unlocking time, user identity, etc.

[0034] When the central management system resumes normal operation, the intelligent lock automatically uploads the emergency unlocking event information stored locally to the central management system. After receiving the emergency unlocking event information, the central management system automatically generates an investigation instruction. For example, the instruction content can be "Please verify the emergency unlocking event of resident XX at XX time" and pushes this instruction to the security management personnel.

[0035] In some specific embodiments, the network monitoring module of the intelligent lock can send a heartbeat packet to the central management system every 30 seconds. If no response from the central management system is received for 10 consecutive heartbeat packets, it is determined that the network connection is interrupted. The system fault signal can be actively sent by the central management system when it detects situations such as server overload and database anomalies on its own. The switching of the emergency mode is achieved by modifying the mode flag bit inside the intelligent lock. The local low-precision face recognition algorithm can be an algorithm with relatively low computational complexity but fast recognition speed, such as the face detection algorithm based on Haar features combined with a simple feature vector comparison method. The emergency unlocking event information can be stored in the Flash memory of the intelligent lock in the form of a log file. The investigation instruction can be encoded in XML format, including fields such as event type, occurrence time, user ID, etc., and is sent to the instruction processing module of the central management system through the message queue middleware.

[0036] Furthermore, the steps of this application for switching to the emergency mode to deactivate the original high-precision face recognition executed in the central management system and enable the low-precision face recognition executed locally on the intelligent lock include: Switch to the emergency mode, and control the communication unit to change the face image information obtained from recognition from being originally sent to the central management information to being sent to the local processing unit; Control the local processing unit of the intelligent lock to load the low-precision algorithm to perform face recognition on the obtained face image information.

[0037] Among them, the switch to the emergency mode can be triggered when the monitoring module detects a network connection interruption or receives a system fault signal.

[0038] The switching process of the emergency mode includes the change of system status flags. For example, a status register is set inside the system. When the emergency condition is met, the value of the status register is modified to the value representing the emergency mode.

[0039] The switching of the data transmission path of the communication unit is achieved by configuring the internal routing table or using a switching circuit. For example, in the normal mode, the routing table is configured with the address for sending face image data to the central management system server; in the emergency mode, the routing table is dynamically modified, and the data sending address is changed to the address of the local processing unit.

[0040] The local processing unit can load the low-precision algorithm in the way of dynamic link library loading, and pre-store the low-precision algorithm library locally. After the emergency mode is activated, the system instructs the local processing unit to call the low-precision algorithm library from the memory and load it into the running memory to provide algorithm support for subsequent face recognition. Compared with the high-precision algorithm, the low-precision algorithm reduces the computational complexity on the premise of ensuring the basic recognition performance, so as to adapt to the computational resource constraints of the local processing unit.

[0041] Specifically, when the smart lock is working normally, it continuously monitors the network connection status and the status signal of the central management system. Once it detects that the network interruption duration exceeds the preset threshold, or receives a fault signal sent by the central management system, it immediately starts the emergency mode switching process.

[0042] First, send an emergency mode switching instruction to the system core control unit, modify the system status flag, and switch the working mode of the smart lock to the emergency mode. Subsequently, control the communication unit, such as the WiFi module or the cellular module, to change its data routing configuration, and switch the face image data collected by the face recognition module from the network interface originally sent to the central management system to the data interface of the local processing unit. At the same time, load the low-precision face recognition algorithm pre-stored in the local memory. After the algorithm is loaded, the local processing unit starts to receive the face image data from the face recognition module and performs face recognition processing using the loaded low-precision algorithm. When the local low-precision face recognition comparison is successful, the execution module receives the recognition success signal and controls the lock body to perform the unlocking action.

[0043] Furthermore, the present application further proposes that the low-precision algorithm loaded by the local processing unit of the smart lock is a personalized algorithm optimized by parameter iteration.

[0044] Among them, the personalized algorithm is implemented through parameter iterative optimization. The parameter iterative optimization process includes the selection of an initial low-precision algorithm and iterative optimization based on the data of successful high-precision face recognition unlocking in the user's daily use. The initial low-precision algorithm can select a face recognition algorithm with relatively low computational complexity, such as an algorithm based on Haar features or LBP features. In daily use, when the user successfully unlocks the lock through high-precision face recognition of the central management system, the system extracts the face feature information of this successful recognition. These information are used to iteratively adjust the parameters of the local low-precision algorithm. The methods of iterative optimization include, but are not limited to, the gradient descent method, etc. Through continuous iterative optimization, the parameters of the low-precision algorithm are gradually adjusted to better adapt to the facial features of the current user. Thus, the personalized algorithm is constructed.

[0045] Specifically, the general low-precision algorithm may have deficiencies in facial feature extraction and matching, and cannot fully take into account the facial characteristics of all users, resulting in a reduced recognition rate for some users during emergency unlocking. However, the personalized algorithm optimized by parameter iteration, due to integrating the facial feature information of the user during the high-precision recognition process, the algorithm parameters can better match the user's facial biometric characteristics. Thus, during emergency unlocking, the personalized algorithm can more accurately recognize the user's face, improving the success rate and reliability of emergency unlocking.

[0046] Furthermore, this application also proposes: After the intelligent lock executes the unlocking instruction through high-precision face recognition performed by the central management system, extract the face feature information recognized by the high-precision face recognition; Perform parameter iterative optimization on the low-precision algorithm according to the face feature information and the recognition passed result to obtain the personalized algorithm.

[0047] Among them, the above solution is that when the intelligent lock is in normal use and the high-precision face recognition through the central management system is successful in unlocking, the system will extract the face feature information used by the high-precision face recognition at this time. Then, the system uses the face feature information of these successful recognitions and the result of successful recognition as feedback to perform parameter iterative optimization on the low-precision face recognition algorithm stored locally in the intelligent lock, making the low-precision algorithm gradually become a personalized algorithm that better conforms to the facial features of the current user. In this way, the low-precision algorithm can continuously learn and adapt to the user's face features, thereby improving the success rate and reliability of local low-precision face recognition during emergency unlocking.

[0048] Specifically, the process of obtaining the personalized algorithm can be described as follows: In the normal working mode of the intelligent lock, every time the user successfully unlocks the lock through high-precision face recognition, the system background collects the face feature data successfully recognized this time. This data can include the key point coordinates of the face, texture information, etc. After collecting a sufficient number of successfully recognized data, for example, collecting the successfully recognized data in the past week or month, the system analyzes this data. The analysis process includes: First, cleaning and preprocessing the face feature data to remove outliers and noise interference; then, using this high-quality face feature data to fine-tune the model parameters of the local low-precision face recognition algorithm. The direction of parameter adjustment is to make the low-precision algorithm more inclined to recognize these face features that have been successfully recognized in subsequent recognition processes. For example, if the low-precision algorithm is based on traditional machine learning algorithms, the decision boundary of the classifier can be adjusted to better adapt to the face feature distribution of the current user. If the low-precision algorithm is based on a deep learning model, the collected face feature data can be used to fine-tune and train the model to update the model parameters. Through this iterative optimization method, the performance of the local low-precision face recognition algorithm can be gradually improved and become more personalized, thereby improving the unlocking success rate in the emergency mode.

[0049] In some specific embodiments, the intelligent lock can be configured to extract and store the face feature information of the current successful recognition every time it successfully unlocks through high-precision face recognition of the central management system. When the stored face feature information reaches a preset number, such as 50 times or 100 times, the intelligent lock automatically starts the parameter iterative optimization program of the local low-precision face recognition algorithm during the idle period, such as at night or during non-frequent use periods of the user. This program first analyzes the stored face feature information, calculates statistical quantities such as the mean and variance of the face features, and evaluates the stability and variability of the face features. Then, according to the analysis results, relevant parameters of the low-precision face recognition algorithm are adjusted, such as adjusting the feature weights, matching thresholds, etc. The adjusted parameters are applied to the local low-precision face recognition algorithm for subsequent emergency unlocking scenarios. As a preferred embodiment, to ensure the effectiveness and security of the optimization process, the parameter iterative optimization program can be designed as an incremental update, that is, only make small adjustments to the model parameters each time and conduct effect evaluation. Only when the optimization effect reaches the preset index, the new parameter configuration is officially applied to the system. In addition, a security verification mechanism can be added to the optimization process, for example, requiring user authorization or system administrator review to avoid malicious programs tampering with the optimization process and affecting system security.

[0050] It should be noted that the parameter iterative optimization of the low-precision algorithm can be implemented on the central management system.

[0051] Furthermore, the personalized algorithm proposed in this application can be implemented based on a neural network.

[0052] Among them, the neural network model structure can be selected as a convolutional neural network, and this network model at least includes a convolutional layer, a pooling layer, and a fully connected layer. The convolutional layer is configured to extract local features in the face image, such as edges, textures, and angles. The pooling layer is configured to reduce the spatial dimension of the feature map, reduce computational complexity, and enhance the translational invariance of the features. The fully connected layer is configured to map the extracted features to the face recognition result. The activation function of the neural network model is selected as the ReLU function to increase the non-linear expression ability of the network and accelerate the convergence speed. The loss function of the neural network model is selected as the cross-entropy loss function, which is used to measure the gap between the model prediction result and the true label. The optimizer is selected as the Adam optimizer, which is used to adaptively adjust the learning rate and accelerate the model training speed.

[0053] Specifically, the personalized algorithm is implemented through a neural network, leveraging the powerful feature learning ability and non-linear mapping ability of the neural network to improve the performance of the low-precision face recognition algorithm. In the emergency unlocking mode, the intelligent lock local processing unit loads the pre-trained neural network model. When the user performs face recognition, the camera captures the face image, and the image is input into the neural network model. The neural network model extracts the features of the face image layer by layer and finally outputs the face recognition result in the fully connected layer. By comparing the output result with the preset threshold, it is determined whether the face recognition is successful. The neural network model can learn and capture the subtle and key features in the face image, and can achieve relatively accurate face recognition even under low-precision recognition conditions. The self-learning ability of the neural network enables the personalized algorithm to continuously optimize and adjust the model parameters to adapt to the facial feature changes of different users, improving the success rate of emergency unlocking and the user experience.

[0054] In some specific embodiments, the training process of the neural network model is performed as follows.

[0055] First, collect the user face image data for unlocking through high-precision face recognition. Then, preprocess the collected face image data, including face detection, alignment, and normalization. Next, according to the constructed convolutional neural network model, initialize the model parameters. After that, use the preprocessed face image data to train the neural network model, continuously adjust the model parameters through the backpropagation algorithm, minimize the loss function, and improve the face recognition accuracy of the model. The trained model is deployed to the intelligent lock local processing unit and used as a personalized algorithm for the emergency unlocking scenario.

[0056] Furthermore, the present application proposes that the personalized algorithm is implemented based on the update of the face template. The steps of parameter iterative optimization of the low-precision algorithm according to the face feature information and the recognition-passed results to obtain the personalized algorithm include: Obtain the face feature information of multiple successful recognitions, and obtain the corresponding face feature vectors for the face feature information of multiple successful recognitions; Calculate the variances of each dimension of the face feature vectors for multiple times, and determine the dimensions with variances lower than the preset threshold as stable features; Perform weighted averaging on the face feature vectors in chronological order, where the latest features obtain higher weights, to form a time-weighted feature vector; Combine the recognized stable features with the time-weighted feature vector to update the original local face template; Based on the importance of the stable features, adjust the weight coefficients of each feature dimension in the low-precision algorithm, and at the same time adjust the matching threshold, so as to obtain the personalized algorithm.

[0057] Among them, obtaining the face feature information of multiple successful recognitions means that after the intelligent lock successfully unlocks through high-precision face recognition, the face feature information is extracted and stored.

[0058] Obtaining the corresponding face feature vectors for the face feature information of multiple successful recognitions means converting the face feature information of each successful recognition into feature vector data that can be processed by a computer.

[0059] Calculating the variances of each dimension of the face feature vectors for multiple times and determining the dimensions with variances lower than the preset threshold as stable features means analyzing the numerical fluctuation degree of each dimension in the multiple feature vectors, and the dimensions with small numerical fluctuations are considered stable features. For example, the distance between the eyes changes little in multiple recognitions and can be regarded as a stable feature.

[0060] Performing weighted averaging on the face feature vectors in chronological order, where the latest features obtain higher weights, to form a time-weighted feature vector means that when integrating the face feature information of multiple times, the feature information at a closer time point is given a higher weight to adapt to the subtle changes in the user's face over time.

[0061] Combining the recognized stable features with the time-weighted feature vector to update the original local face template means fusing the stable features and the time-weighted feature vector to optimize and update the face template stored locally in the intelligent lock, making the template more personalized and timely.

[0062] Based on the importance of the stable features, adjusting the weight coefficients of each feature dimension in the low-precision algorithm and at the same time adjusting the matching threshold to obtain the personalized algorithm means that in the low-precision face recognition algorithm, the weights of the stable features are increased, and at the same time the matching threshold is adjusted to improve the recognition performance of the low-precision algorithm.

[0063] Specifically, in the high-precision face recognition and unlocking process, after verification is passed and the door lock is opened, the system background runs the feature extraction module. This module extracts face feature information from the successfully recognized face image this time and converts it into feature vectors of a fixed dimension. These feature vectors are stored in the local database, associated with timestamps, forming a time series of face feature vectors.

[0064] The data analysis module of the system regularly analyzes the time series of face feature vectors. For each feature dimension, it calculates the variance on the time series. A preset variance threshold is used to judge the stability of the feature dimension. For example, if the dimension variance is lower than 0.01, then this dimension is determined to be a stable feature.

[0065] The time-weighted average module assigns weights to each face feature vector according to a time decay function. The closer the time, the higher the weight. For example, the weight of the latest feature vector is 0.8, the second latest is 0.7, and so on. After weighted averaging, a time-weighted feature vector is formed, reflecting the average state of recent face features.

[0066] The template update module receives the stable feature information and the time-weighted feature vector. The stable features are used to guide the direction of template update, and the time-weighted feature vector is used as the basis for update. The update process integrates stable features and time-weighted features. For example, the mean value of stable features is replaced by the time-weighted feature vector, and unstable features remain their original values. The algorithm adjustment module analyzes the importance of stable features in face recognition and adjusts the weight coefficients of each feature dimension in the low-precision algorithm. The weight coefficients of stable features are increased, and the weight coefficients of unstable features are decreased. The matching threshold is also adjusted according to the weight of stable features. For example, after the weight of stable features is increased, the matching threshold can be appropriately reduced to improve the recognition sensitivity. Through the above steps, the low-precision face recognition algorithm is personalized adjusted to better adapt to the facial features of specific users.

[0067] In some specific embodiments, when a user uses the smart lock for the first time, the first three successful unlocks are achieved through high-precision face recognition by the central management system. After each successful unlock, the smart lock extracts the face feature information of Zhang San and obtains three multi-dimensional feature vectors V1, V2, and V3. The system calculates the variances of each dimension of these three vectors. Assuming that the variances of dimensions D5, D12, and D87 are lower than the preset threshold, they are determined as stable features. The time-weighted average module performs weighted averaging on V1, V2, and V3 with weights of 0.6, 0.8, and 1.0 respectively to obtain the time-weighted feature vector Vt. The template update module replaces the dimension values corresponding to the stable features D5, D12, and D87 in the local low-precision algorithm's face template with the corresponding dimension values in Vt. The algorithm adjustment module increases the weight coefficients of the three dimensions D5, D12, and D87 in the low-precision algorithm to 1.2, and keeps the weight coefficients of the remaining dimensions unchanged. The matching threshold is adjusted from 0.5 to 0.45. Thus, a personalized low-precision face recognition algorithm customized for the user is generated. When the central management system fails and the smart lock switches to the emergency mode, this personalized algorithm is enabled to improve the user's unlock success rate in case of emergency.

[0068] Furthermore, the present application proposes to classify emergency events; determine the information type to be uploaded to the central management system for the local stored emergency unlock event information according to the classification of the emergency events; and generate an investigation instruction according to the uploaded information type and the emergency event classification.

[0069] Among them, classifying emergency events can be based on environmental parameters during emergency unlocking, such as unlocking time, duration, number of attempts, etc., for preliminary classification, and the emergency events are divided into three categories: regular emergency, suspicious emergency, and high-risk emergency.

[0070] Specifically, a regular emergency refers to an emergency unlock event that occurs under an expected behavior pattern. A suspicious emergency refers to an emergency unlock event with certain abnormal characteristics but not sufficient to be determined as high-risk. A high-risk emergency refers to an emergency unlock event with obvious abnormal characteristics, which may indicate a security threat. The levels of investigation instructions required for different emergency types can be different.

[0071] Determining the information type according to the emergency event classification means determining the content of the emergency unlock event information that needs to be uploaded to the central management system for different types of emergency events. For example, for a regular emergency, only basic information is uploaded to generate a simple report; for a suspicious emergency, the complete unlocking process data and low-precision recognition matching degree data are uploaded to generate a detailed report; for a high-risk emergency, in addition to the above data, environmental data and household activity data within a short time before and after unlocking are also uploaded.

[0072] For system failure events, the information type may include information such as unlocking time, user identity information, system failure code, and system log when the failure occurs.

[0073] Generating an investigation instruction means generating specific investigation instructions based on the determined information type and emergency event classification for investigating and handling the current emergency unlocking event. For example, different levels of investigation instructions are automatically generated based on the emergency type. For example, a high-risk emergency automatically triggers a property security patrol and sends a notice to the designated contact of the owner. Whether manual review and intervention are required is determined according to the abnormal degree of the unlocking scenario.

[0074] Furthermore, when the central management system resumes normal operation, the steps of uploading the emergency unlocking event information stored locally to the central management system and generating an investigation instruction include: when performing low-precision face recognition for emergency unlocking, recording the timestamp of the unlocking operation and the user identity information recognized; generating a query instruction for investigating the geographical location of the corresponding household user at the current timestamp; after uploading the emergency unlocking event information stored locally when the central management system resumes normal operation, performing spatio-temporal consistency verification, including: querying the geographical location of the corresponding household user at the corresponding timestamp according to the query instruction; calculating the physical distance and possible moving time between the geographical location of the corresponding household user and the residential location; judging spatio-temporal consistency according to the physical distance and possible moving time. When spatio-temporal inconsistency is detected, generating a high-priority investigation instruction and marking it as a potential security risk event; sending a security alert notice to the household head.

[0075] Among them, when the intelligent lock completes low-precision face recognition unlocking in the emergency mode, the local processing unit of the intelligent lock records the exact time point when the unlocking operation occurs and the user identity information recognized by the low-precision face recognition algorithm.

[0076] The emergency unlocking event information is stored in the local memory of the intelligent lock.

[0077] The system generates a geographical location query instruction for querying the geographical location of the corresponding household user at the unlocking moment after the central management system resumes. The geographical location query instruction can take various forms. For example, a query request based on the positioning record of the user's mobile device. When the central management system resumes normal operation, the intelligent lock uploads the emergency unlocking event information stored locally, including the timestamp, user identity information, etc. to the central management system. After receiving the emergency unlocking event information, the central management system performs spatio-temporal consistency verification. The system sends a request according to the previously generated query instruction to obtain the geographical location of the user corresponding to the unlocking timestamp.

[0078] The system calculates the physical distance between the geographical location of the corresponding household user when unlocking the door and the location of the family residence. Based on the calculated physical distance, it estimates the possible shortest time required for the user to move from this geographical location to the residence, and compares the estimated travel time with the time difference from the moment the user is located to the actual unlocking moment, so as to judge the spatio-temporal consistency.

[0079] If the geographical location of the user and the location of the residence are unreasonable in terms of time and space. For example, when unlocking the door, the geographical location of the user is very far from the residence and cannot be reached within a short time, it is determined as spatio-temporal inconsistency. When spatio-temporal inconsistency is detected, the system generates a high-priority investigation instruction, instructing the security personnel to conduct a key verification of this emergency unlocking event, and marks this event as a potential security risk event for subsequent tracking and analysis. It can automatically send a security alert notification to the residential household owner, prompting the owner that there may be a security risk and suggesting that the owner pay attention to the security status of the residence.

[0080] Specifically, in the security system, when the central management system fails, the intelligent lock switches to the emergency mode and uses the local low-precision face recognition algorithm to unlock the door. In order to address the problem of insufficient security risk assessment that may exist after the system resumes operation, this solution adds a spatio-temporal consistency verification link after the central management system resumes normal operation.

[0081] After the household successfully unlocks the door using low-precision face recognition, the intelligent lock records the unlocking timestamp and the identified user identity information. The system automatically generates a geographical location query instruction to query the geographical location of the user at the unlocking time point. The geographical location information can be obtained from the location service of the mobile operator or the location record of the application. After the central management system resumes, the intelligent lock uploads the emergency unlocking event information, and the central management system performs spatio-temporal consistency verification.

[0082] First, the system obtains the geographical location of the user at the unlocking time point according to the query instruction. Then, the system calculates the distance between the user's geographical location at that time and the location of the residence, and estimates the possible travel time. By comparing the distance and the travel time, the system judges whether it is spatio-temporally logical for the user to appear at the residence during emergency unlocking. For example, if the user uses emergency unlocking at 3 am, but the location information shows that the user is in a city 500 kilometers away, the system will determine it as spatio-temporal inconsistency. When spatio-temporal inconsistency occurs, the system determines that there is a potential security risk in this emergency unlocking event, generates a high-priority investigation instruction, and marks it as a potential security risk event. At the same time, it sends a security alert notification to the household owner. Through spatio-temporal consistency verification, the system can more effectively identify and warn of potential illegal intrusions or security vulnerabilities, improving the security of the intelligent lock system.

[0083] Further, the steps proposed in this application for switching to the emergency mode to deactivate the high-precision face recognition originally executed in the central management system and activate the low-precision face recognition executed locally on the smart lock include: Obtain the type of fault; According to the type of fault, adaptively switch to the emergency mode to deactivate the high-precision face recognition originally executed in the central management system and activate the low-precision face recognition executed locally on the smart lock, where the type of fault includes but is not limited to: network communication interruption, server response timeout, system operation anomaly; For the fault of network communication interruption, select an algorithm configuration with lower computational complexity but more feature points; For the fault of server response timeout, select an algorithm configuration with medium computational complexity and set a short recognition timeout threshold; For the fault of system operation anomaly, select the algorithm configuration with the highest security level.

[0084] Among them, obtaining the type of fault can be executed by the smart lock. The monitoring module continuously monitors the network connection status between the smart lock and the central management system.

[0085] The monitoring of the network connection status can be achieved by periodically sending heartbeat packets and waiting for a response. If no response from the central management system is received within the preset time, it is determined as a network communication interruption fault.

[0086] The server response timeout fault can be determined if the smart lock does not receive a response from the server within the set time after sending a request to the central management system.

[0087] The detection of the system operation anomaly fault can rely on the status signal sent by the central management system. The central management system can monitor its own operation status in real time and actively send a system fault signal to the smart lock when detecting a system anomaly.

[0088] Adapting to switch to the emergency mode according to the type of fault means that after the switching module of the smart lock receives the fault type information sent by the monitoring module, it automatically adjusts the configuration of the face recognition algorithm in the emergency mode according to the preset correspondence between the type of fault and the algorithm configuration.

[0089] For the network communication interruption fault, the switching module controls the smart lock to activate a local face recognition algorithm with lower computational complexity but more feature points. For example, a face recognition algorithm based on a lightweight convolutional neural network can be selected. Such an algorithm reduces the consumption of computing resources and improves the recognition speed while ensuring a certain recognition accuracy.

[0090] For a server response timeout failure, the switching module controls the smart lock to enable a local face recognition algorithm with medium computational complexity and sets a short recognition timeout threshold. For example, an algorithm with a slightly higher computational complexity but higher recognition accuracy than the aforementioned algorithm can be selected, and the recognition timeout threshold can be set to, for example, 2 seconds to avoid long waiting times for users while ensuring a certain level of recognition accuracy.

[0091] For a system operation anomaly failure, the switching module controls the smart lock to enable the local face recognition algorithm with the highest security level. For example, a high-precision face recognition algorithm based on a deep convolutional neural network can be selected. Although such algorithms have a relatively high computational complexity and a relatively slow recognition speed, they are more secure in terms of security.

[0092] Specifically, for a network communication interruption failure, the algorithm configuration with lower computational complexity but more feature points needs to be selected because when the network is completely interrupted, the smart lock must rely entirely on the local processor to run, and the processing capacity of edge devices is limited. At this time, how to maintain sufficient recognition accuracy under severely limited computing resources becomes a problem. To this end, the overall complexity of the algorithm can be reduced to adapt to the local processing capacity, and the number of feature points can be increased as a compensation to maintain the recognition performance.

[0093] For a server response timeout failure, the algorithm configuration with medium computational complexity and a short recognition timeout threshold needs to be selected because a server response timeout indicates that the network connection exists but is unstable, and the system needs to balance between waiting for a cloud response and local processing. At this time, how to handle the system response delay caused by network connection uncertainty becomes a problem. To this end, a medium-complexity algorithm can be used to maintain good recognition performance, and a short timeout threshold can be set to avoid the system waiting for an uncertain response for a long time.

[0094] For a system operation anomaly failure, the algorithm configuration with the highest security level needs to be selected because a system anomaly may imply potential security threats or attack attempts, and at this time, the security priority should be higher than convenience. At this time, how to minimize the risk of unauthorized access in the case of a system anomaly becomes a problem. To this end, the algorithm configuration with the highest security level can be enabled, and the possible increase in the false rejection rate can be accepted in exchange for higher security.

[0095] This differential configuration strategy for different failure types solves the balance problem among resource utilization, response time, and security protection of the smart lock in abnormal situations, and is an intelligent adaptive mechanism based on the failure scenario.

[0096] In the solution of this application, after the low-precision algorithm is optimized into a personalized algorithm through parameter iteration, it can be further divided into adaptive configurations according to different failure types. This division aims to balance the computing resources, response speed, and security requirements, specifically including: The network communication interruption adaptation algorithm can be based on the original low-precision algorithm, select to reduce the calculation level, and at the same time retain more feature points (such as facial key point markings) to maintain the basic recognition accuracy in a local environment with limited computing power. This configuration prioritizes processing speed and resource efficiency and is suitable for scenarios that rely entirely on local processing.

[0097] The server response timeout adaptation algorithm can be based on the original low-precision algorithm, maintain medium computational complexity, but add a timeout control mechanism and set a shorter recognition time limit (such as 2 seconds). This configuration can adopt a phased recognition strategy to balance recognition quality and response speed within a limited time and is suitable for scenarios where the network connection is unstable but may still recover.

[0098] The system operation exception adaptation algorithm can be based on the original low-precision algorithm, enhance the security verification mechanism, and can increase the matching threshold or add additional verification steps. This configuration accepts higher computational overhead and longer response times in exchange for the highest level of security and is suitable for scenarios that may face security threats.

[0099] Through this algorithm adaptation mechanism based on the type of failure, the intelligent lock can intelligently adjust the local recognition strategy in the abnormal state of the system to achieve the optimal configuration of the emergency unlocking function.

[0100] Specifically, it can refer to Figure 3 , in some specific embodiments, the solution of the present application is specifically as Figure 3 shown. The central management system is located at the top layer of the system and is responsible for high-precision face recognition and access control. Under normal circumstances, the face data of the user will be transmitted to the central management system for high-precision recognition. After passing the verification, an unlocking instruction will be sent to the intelligent lock. At the same time, the central system is also responsible for providing a parameter optimization data stream to the personalized algorithm to continuously improve the effect of the local algorithm.

[0101] Monitoring and switching is responsible for monitoring the network connection status between the intelligent lock and the central management system and receiving system status signals. When a network interruption is detected or a system failure signal is received, it will automatically switch to the emergency mode. Specifically, in the case of a network interruption, an algorithm configuration with lower computational complexity but more feature points can be selected. For server response timeout failures, it is necessary to select an algorithm configuration with medium computational complexity. In the case of a system failure, a high-security-level algorithm configuration is selected.

[0102] The personalized low-precision algorithm is implemented based on face template updates. Through the parameter optimization data stream obtained from the high-precision face recognition system, the local low-precision recognition algorithm is continuously improved. According to the facial features of the user, a personalized face recognition algorithm suitable for a specific user is generated, reducing the computational resource requirements while ensuring the recognition accuracy.

[0103] The intelligent lock performs low-precision face recognition comparison and unlocking operations locally. In the emergency mode, it receives a switching instruction, simultaneously loads an optimized low-precision personalized algorithm, performs local face recognition and unlocking operations, and stores the emergency unlocking event information locally.

[0104] Post-event processing and security verification, classifying, verifying, and providing security responses for emergency unlocking events. The events are divided into regular emergencies, suspicious emergencies, and high-risk emergencies. By querying the user's geographical location, calculating the distance from the residence, and the possible travel time, the rationality of the unlocking behavior is verified. When a suspicious situation is detected, a high-priority investigation instruction is generated and a security alert is sent.

[0105] This design ensures that when the central management system fails, the intelligent lock can still operate normally through the local low-precision face recognition algorithm, while ensuring sufficient security and post-event management tracking capabilities.

[0106] In a second aspect, referring to Figure 2 , the present application further proposes an unlocking mode switching device for an intelligent lock, including: A monitoring module 210 for monitoring the network connection status between the intelligent lock and the central management system, and receiving status signals from the central management system; A switching module 220 for switching to the emergency mode to deactivate the high-precision face recognition originally performed in the central management system and enable the low-precision face recognition performed locally on the intelligent lock when the network connection is interrupted for more than a preset duration threshold or when a system failure signal sent by the central management system is received; An execution module 230 for executing the unlocking instruction when the low-precision face recognition performed locally on the intelligent lock is successfully compared, and storing the emergency unlocking event information locally on the intelligent lock after the intelligent lock is opened; A post-processing module 240 for uploading the locally stored emergency unlocking event information to the central management system and generating an investigation instruction when the central management system resumes normal operation.

[0107] In addition, in some preferred embodiments, the unlocking mode switching device for an intelligent lock proposed by the present application can perform any one of the steps in the above method.

[0108] By switching to the local low-precision face recognition mode when the network connection is interrupted or the system fails, the problem that the intelligent lock cannot be normally unlocked when the central management system fails is solved, and the effect of ensuring that the household can quickly and effectively open the door lock in the scenario where the central management system fails or the network is interrupted is achieved.

[0109] The above are only embodiments of the present application and are not intended to limit the protection scope of the present application. For those skilled in the art, various modifications and changes can be made to the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included within the protection scope of the present application.

Claims

1. A method for switching the unlocking mode of an intelligent lock, the method comprising: Monitoring the network connection status between the intelligent lock and the central management system, and receiving a status signal from the central management system; When the network connection is interrupted for more than a preset duration threshold, or a system failure signal sent by the central management system is received, switching to an emergency mode to deactivate the high-precision face recognition originally performed in the central management system and enabling the low-precision face recognition performed locally on the intelligent lock; When the low-precision face recognition performed locally on the intelligent lock is successfully matched, executing an unlocking instruction, and after the intelligent lock is opened, storing emergency unlocking event information locally on the intelligent lock; When the central management system resumes normal operation, uploading the locally stored emergency unlocking event information to the central management system and generating an investigation instruction.

2. The unlocking mode switching method of an intelligent lock according to claim 1, characterized in that, The step of switching to the emergency mode to deactivate the high-precision face recognition originally performed in the central management system and enabling the low-precision face recognition performed locally on the intelligent lock includes: Switching to the emergency mode, and controlling the communication unit to switch the recognized face image information from originally being sent to the central management information to being sent to the local processing unit; Controlling the local processing unit of the intelligent lock to load a low-precision algorithm to perform face recognition on the obtained face image information.

3. The unlocking mode switching method of an intelligent lock according to claim 2, characterized in that, The low-precision algorithm loaded by the local processing unit of the intelligent lock is a personalized algorithm optimized by parameter iteration.

4. The unlocking mode switching method of an intelligent lock according to claim 3, wherein, It further includes: After the intelligent lock executes an unlocking instruction through the high-precision face recognition performed by the central management system, extracting the face feature information recognized by the high-precision face recognition; Performing parameter iteration optimization on the low-precision algorithm according to the face feature information and the recognition passed result to obtain a personalized algorithm.

5. The unlocking mode switching method of an intelligent lock according to claim 4, wherein, The personalized algorithm is implemented based on a neural network.

6. The unlocking mode switching method of an intelligent lock according to claim 4, wherein, The personalized algorithm is implemented based on face template update. The step of performing parameter iteration optimization on the low-precision algorithm according to the face feature information and the recognition passed result to obtain a personalized algorithm includes: Obtaining the face feature information of multiple successful recognitions, and obtaining the corresponding face feature vectors for the face feature information of multiple successful recognitions; Calculating the variances of each dimension of the multiple face feature vectors, and determining the dimensions with variances lower than the preset threshold as stable features; Performing weighted averaging on the face feature vectors in chronological order, where the latest features obtain higher weights, to form a time-weighted feature vector; Combining the recognized stable features with the time-weighted feature vector to update the original local face template; Based on the importance of the stable features, adjusting the weight coefficients of each feature dimension in the low-precision algorithm and simultaneously adjusting the matching threshold to obtain a personalized algorithm.

7. A method for switching the unlocking mode of an intelligent lock according to claim 1, characterized in that, The step of, when the central management system resumes normal operation, uploading the locally stored emergency unlocking event information to the central management system and generating an investigation instruction includes: Classifying the emergency events; Determining the information type for uploading the locally stored emergency unlocking event information to the central management system according to the classification of the emergency events; Generating an investigation instruction according to the uploaded information type and the classification of the emergency events.

8. The unlocking mode switching method of an intelligent lock according to claim 1, characterized in that The step of, when the central management system resumes normal operation, uploading the locally stored emergency unlocking event information to the central management system and generating an investigation instruction includes: When performing emergency unlocking for low-precision face recognition, record the timestamp of the unlocking operation and the recognized user identity information; Generate a query instruction to investigate the geographical location of the corresponding household user at the current timestamp; After the central management system resumes normal operation, after uploading the local stored emergency unlocking event information, also perform spatio-temporal consistency verification, including: Query the geographical location of the corresponding household user at the corresponding timestamp according to the query instruction; Calculate the physical distance and possible travel time between the geographical location of the corresponding household user and the residential location; Judge spatio-temporal consistency based on the physical distance and possible travel time. When spatio-temporal inconsistency is detected, generate a high-priority investigation instruction and mark it as a potential security risk event; Send a security alert notification to the household head.

9. The unlocking mode switching method of an intelligent lock according to claim 1, characterized in that, The step of switching to the emergency mode to deactivate the high-precision face recognition originally performed in the central management system and activate the low-precision face recognition performed locally on the smart lock includes: Obtain the fault type; Perform adaptive switching to the emergency mode according to the fault type to deactivate the high-precision face recognition originally performed in the central management system and activate the low-precision face recognition performed locally on the smart lock, where the fault type includes but is not limited to: network communication interruption, server response timeout, system operation anomaly; For the fault of network communication interruption, select an algorithm configuration with lower computational complexity but more feature points; For the fault of server response timeout, select an algorithm configuration with medium computational complexity and set a short recognition timeout threshold; For the fault of system operation anomaly, select the algorithm configuration with the highest security level.

10. An unlocking mode switching device for an intelligent lock, characterized in that, Include: A monitoring module for monitoring the network connection status between the smart lock and the central management system, and receiving status signals from the central management system; A switching module for switching to the emergency mode to deactivate the high-precision face recognition originally performed in the central management system and activate the low-precision face recognition performed locally on the smart lock when the network connection is interrupted for more than a preset duration threshold or a system fault signal sent by the central management system is received; An execution module for executing the unlocking instruction when the low-precision face recognition performed locally on the smart lock is successfully matched, and storing the emergency unlocking event information locally on the smart lock after the smart lock is opened; A post-processing module for uploading the locally stored emergency unlocking event information to the central management system and generating an investigation instruction after the central management system resumes normal operation.

Images