A secure unlocking method, system, and storage medium based on the linkage between dual door locks and a camera.

Abstract

This invention relates to the field of smart door lock technology, and provides a secure unlocking method, system and storage medium based on the linkage of dual door locks and camera. By establishing a binding relationship between the camera and dual smart door locks, the camera pre-collects facial features and performs a dual comparison with the real-time verification results of the door locks, realizing automatic collaborative unlocking of dual door locks and multi-stage security verification. It has the advantages of improving the efficiency of collaborative unlocking of dual door locks and the reliability of security verification.

Description

Technical Field

[0001] This invention relates to the field of smart door lock technology, and more specifically, to a secure unlocking method, system, and storage medium based on the linkage of dual door locks and a camera. Background Technology

[0002] Important properties such as villas and office areas often have two smart locks installed on the main entrance. While this design effectively improves security, it also brings many inconveniences. For intruders, both locks need to be opened simultaneously to gain entry, increasing the difficulty and time cost of forced entry or spoofed attacks. However, for legitimate users, this design also creates difficulties, requiring them to unlock both locks separately to open the door, a cumbersome and time-consuming process. Current technology lacks a solution that balances security and convenience, failing to achieve intelligent collaborative unlocking of two locks. Furthermore, traditional lock systems and monitoring systems often operate independently, lacking integrated verification, resulting in single security verification steps and insufficient verification information. When a user moves from the monitored area to the main entrance, the system cannot establish a complete identity verification chain, leading to security vulnerabilities.

[0003] To address the aforementioned issues, existing technologies urgently need improvement. Summary of the Invention

[0004] The purpose of this application is to provide a secure unlocking method, system, and storage medium based on the linkage of dual locks and a camera, which has the advantages of improving the efficiency of collaborative unlocking of dual locks and the reliability of security verification.

[0005] A first aspect of the present invention provides a secure unlocking method based on the linkage of dual door locks and a camera, applied to a gate equipped with a first smart door lock and a second smart door lock; the camera is installed on the road leading to the gate; the method includes:

[0006] The system connects the first smart lock, the second smart lock, and the camera to the cloud platform network and registers them with the cloud platform; it also sets the first smart lock, the second smart lock, and the camera to a bound relationship.

[0007] In response to a person passing by the camera, the system captures a facial image based on the camera and determines a first facial feature vector, then sends a first notification message to the first smart lock and the second smart lock; the first notification message includes the camera ID, a list of the first facial feature vectors, and a detection timestamp;

[0008] In response to personnel arriving at the gate, face detection is performed through the first smart door lock to obtain the second face feature vector. After verification by comparison with the local face template, it is determined whether the second face feature vector exists in the first face feature vector list contained in the first notification message, and a first determination result is generated.

[0009] Based on the first determination result, the first smart door lock is controlled to unlock, and at the same time, a second notification message is sent to the second smart door lock; the second notification message includes a second facial feature vector;

[0010] The second smart door lock is controlled to compare the second facial feature vector with the local facial template, and the second smart door lock is unlocked according to the comparison result.

[0011] Furthermore, the cloud platform records door lock information; the door lock information includes a first smart lock ID, a second smart lock ID, a first smart lock IP, a second smart lock IP, a camera ID, and a camera IP.

[0012] The camera records camera binding information; the camera binding information includes camera ID, camera IP, a list of bound door lock IDs, and a list of bound door lock IPs; wherein, the list of bound door lock IDs includes a first smart door lock ID and a second smart door lock ID; the list of bound door lock IPs includes a first smart door lock IP and a second smart door lock IP.

[0013] Furthermore, in response to a person passing by the camera, a facial image is captured based on the camera and a first facial feature vector is determined, and a first notification message is sent to the first smart lock and the second smart lock, including:

[0014] The first smart lock and the second smart lock cache the first notification message for a preset time period; if the preset time period is exceeded, the first smart lock and the second smart lock delete the first notification message from the lock cache.

[0015] Furthermore, in response to a person arriving at the gate, face detection is performed through the first smart door lock to obtain a second face feature vector. After verification by comparing it with a local face template, it is determined whether the second face feature vector exists in the first face feature vector list contained in the first notification message, and a first determination result is generated, including:

[0016] In response to a person arriving at the gate, the first smart door lock acquires the second facial feature vector and compares it with the locally stored facial template; if the comparison is successful, it determines whether the second facial feature vector exists in the first facial feature vector list contained in the first notification message.

[0017] Furthermore, the method also includes:

[0018] In response to a person passing by the camera, the camera captures a facial image and calculates a first facial feature vector, generating a new record in a locally stored table of passing personnel information; wherein each record in the table of passing personnel information includes the first facial feature vector and a detection timestamp;

[0019] In response to a person arriving at the gate, the system performs face detection via the first smart lock, obtains a second face feature vector, calculates the estimated time point P after passing the camera, and sends a person confirmation request to the camera. The person confirmation request includes the face feature vector to be confirmed and the estimated time point P after passing the camera. The estimated time point P after passing the camera is the time taken for the first smart lock to perform face detection minus a preset duration. The preset duration is based on the time required to walk from the camera to the gate.

[0020] Based on the personnel confirmation request information, the camera searches the personnel information table for records that match the facial feature vector of the person to be confirmed; if a matching record exists, it verifies whether the detection timestamp of the record is within the preset range of the estimated time point P, and returns a response status to the first smart door lock; the response status includes hit and miss; wherein, if the verification passes, it is marked as hit, otherwise it is marked as miss;

[0021] The first smart lock unlocks according to the response status and sends a second notification message to the second smart lock; the second notification message includes a second facial feature vector.

[0022] The second smart door lock is controlled to compare the second facial feature vector with the local facial template, and the second smart door lock is unlocked according to the comparison result.

[0023] Furthermore, the gate also includes one or more third smart locks; the first smart lock is the main lock, equipped with a PIR sensor; the other locks are secondary locks; and both the main lock and secondary locks store local face templates; the main lock and secondary locks each use different face recognition models; the method further includes:

[0024] The master lock performs facial recognition on the user, generates a first recognition result, and randomly selects any secondary lock to send a facial recognition message carrying the first recognition result;

[0025] The secondary lock receives a face recognition message sent by the primary lock; the face recognition message contains a set of recognition results from both the primary and secondary locks.

[0026] If the number of successful recognitions in the recognition result set exceeds a preset threshold, then all smart locks are controlled to perform unlocking operations; otherwise, any other secondary lock that has not participated in the recognition is randomly selected to send a message carrying face recognition until the number of successful recognitions in the recognition result set exceeds the preset threshold or all locks have participated in the recognition.

[0027] Secondly, this embodiment also proposes a security unlocking system based on the linkage of dual door locks and a camera, applied to a gate equipped with a first smart door lock and a second smart door lock; the camera is installed on the road leading to the gate; the system includes:

[0028] The registration module is used to connect the first smart lock, the second smart lock, and the camera to the cloud platform network and register them with the cloud platform; and to set the first smart lock, the second smart lock, and the camera as bound together.

[0029] The first notification module is used to respond to a person passing by the camera, capture a face image based on the camera and determine a first face feature vector, and send a first notification message to the first smart lock and the second smart lock; the first notification message includes the camera ID, the first face feature vector list and the detection timestamp;

[0030] The generation module is used to respond to the arrival of personnel at the gate, perform face detection through the first smart door lock, obtain the second face feature vector, and after the verification is passed by comparing with the local face template, determine whether the second face feature vector exists in the first face feature vector list contained in the first notification message, and generate a first determination result.

[0031] The second notification module is used to control the first smart lock to unlock based on the first determination result, and at the same time send a second notification message to the second smart lock; the second notification message includes a second facial feature vector;

[0032] The unlocking control module is used to control the second smart door lock to perform a comparison between the second facial feature vector and the local facial template, and to perform the unlocking of the second smart door lock based on the comparison result.

[0033] The cloud platform records door lock information; the door lock information includes a first smart lock ID, a second smart lock ID, a first smart lock IP, a second smart lock IP, a camera ID, and a camera IP.

[0034] The camera records camera binding information; the camera binding information includes camera ID, camera IP, a list of bound door lock IDs, and a list of bound door lock IPs; wherein, the list of bound door lock IDs includes a first smart door lock ID and a second smart door lock ID; the list of bound door lock IPs includes a first smart door lock IP and a second smart door lock IP.

[0035] Thirdly, in this embodiment, a storage medium is also provided, which stores a computer program; the program is loaded and executed by a processor to implement the steps of the secure unlocking method based on the linkage between dual door locks and a camera as described in the first aspect above.

[0036] As can be seen from the above, the security unlocking method, system and storage medium based on the linkage of dual door locks and camera provided in this application establishes a binding relationship between the camera and dual smart door locks, and uses the camera to pre-collect facial features and perform dual comparison with the real-time verification results of the door locks, thereby realizing automatic collaborative unlocking of dual door locks and multi-stage security verification. It has the advantages of improving the efficiency of collaborative unlocking of dual door locks and the reliability of security verification. Attached Figure Description

[0037] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart illustrating the steps of a secure unlocking method based on the linkage of a double-door lock and a camera, as disclosed in an embodiment of the present invention.

[0039] Figure 2 This is a schematic diagram of the security unlocking system based on the linkage between a double door lock and a camera, as disclosed in an embodiment of the present invention. Detailed Implementation

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.

[0041] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0042] It should be noted that "multiple" as mentioned in this article refers to two or more.

[0043] In existing technologies, important houses typically employ a dual-lock design to enhance security. However, authorized users must operate both locks separately to gain entry, a cumbersome process. Traditional solutions involve two locks operating independently, lacking a collaborative verification mechanism between devices, forcing users to perform the complete facial recognition process twice. Even if an intruder brute-forces a single lock, security risks remain. While independent dual-lock verification improves security, it sacrifices user convenience. For example, when two smart locks are installed on a villa's front door, users must perform facial recognition on two different locks, increasing operation time and complexity.

[0044] To address the aforementioned issues, a data-sharing mechanism between devices is needed, enabling dual locks to perform collaborative operations based on the same verification information. Analysis revealed that pre-collecting biometric data during the user's approach and pushing it to the lock system can reduce computational latency during on-site verification. Further consideration is given to deploying cameras along the path as pre-verification nodes, splitting the facial recognition process into two stages: pre-collection and real-time verification. This retains the security advantages of independent dual-lock verification while avoiding repetitive operations. The key lies in establishing a binding relationship between the cameras and the dual locks, achieving timely transmission of feature data through a cloud platform, so that the comparison process of the second lock is automatically triggered after the first lock's verification is successful.

[0045] Therefore, this application proposes a secure unlocking method based on the linkage of dual door locks and a camera, applicable to a gate equipped with a first smart door lock and a second smart door lock, with a camera installed on the path leading to the gate. The method includes:

[0046] S101, connect the first smart lock, the second smart lock, the camera to the cloud platform network and register with the cloud platform; set the first smart lock, the second smart lock, and the camera to a binding relationship;

[0047] S102, in response to a person passing by the camera, a face image is captured based on the camera and a first face feature vector is determined, and a first notification message is sent to the first smart lock and the second smart lock; the first notification message includes the camera ID, a list of the first face feature vectors and a detection timestamp.

[0048] S103, in response to personnel arriving at the gate, face detection is performed through the first smart door lock to obtain the second face feature vector. After the verification is passed by comparing with the local face template, it is determined whether the second face feature vector exists in the first face feature vector list contained in the first notification message, and a first determination result is generated.

[0049] S104, based on the first determination result, control the first smart door lock to unlock, and at the same time send a second notification message to the second smart door lock; the second notification message includes a second facial feature vector;

[0050] S105, control the second smart door lock to compare the second face feature vector with the local face template, and unlock the second smart door lock according to the comparison result.

[0051] The smart door lock and camera are bound together via a cloud platform. When a person passes by the camera, the first facial feature vector is calculated and pushed to both door locks. When the person arrives at the door, the first door lock obtains the second facial feature vector and compares it with a pre-stored list. After verification, the first door lock is opened, and the second feature vector is sent to the second door lock for local template comparison to control the second door lock to open.

[0052] The cloud platform network connection refers to establishing communication links between devices via the Internet Protocol (IP). Specifically, the MQTT protocol can be used for device registration and message transmission to ensure reliable synchronization of the binding relationship between the camera and the double-door lock. The camera ID included in the first notification message is used to identify the source of the feature data. UUID encoding can be used to ensure unique device identification and prevent unauthorized devices from forging verification information. The detection timestamp is recorded using an international standard time format, specifically through network time protocol synchronization of the device clock, to limit the effective time range of the feature data and prevent malicious reuse of historical data. The second facial feature vector in the second notification message is transmitted encrypted. Specifically, the AES-256 algorithm can be used to encapsulate the feature data to ensure data security during transmission.

[0053] Specifically, when a person enters the camera-monitored area, the device automatically captures a facial image and extracts feature vectors, generating a verification data packet containing a timestamp and sending it to the dual-lock system. When the first lock detects the user approaching, it first performs local facial verification. If successful, it immediately retrieves a pre-stored list of camera features for rapid comparison. If a match is found, the lock opens directly and forwards the currently collected real-time feature vectors to the second lock. Upon receiving the data, the second lock performs secondary verification using a locally stored template, ensuring the real-time consistency of biometric features. This process, through the superposition of pre-verification and real-time verification, forms a cascading verification process between the two locks, shortening user waiting time while maintaining the security strength of dual verification.

[0054] Compared to existing technologies, traditional dual-lock solutions require users to perform two separate facial recognition operations. This solution, however, uses pre-collected feature data from a camera to enable shared verification information between the two locks, allowing users to trigger the dual-lock linkage with just one recognition. Existing technologies involve completely independent dual-lock verification processes, failing to utilize pre-verification data provided by path monitoring devices, resulting in low verification efficiency. This solution establishes a data channel between devices, transforming path monitoring information into a predictive basis for door lock verification, optimizing the operation process without compromising security.

[0055] Through the above technical solution, this application achieves automated collaborative operation of the dual-lock system. Authorized users only need to perform facial recognition once to unlock both locks, reducing operation time by approximately 50%. Simultaneously, it maintains the security mechanism of independent dual-lock verification; intruders must simultaneously crack the verification systems of both locks to gain unauthorized entry, achieving a security level comparable to traditional dual-lock solutions. Encrypted data transmission and timeliness control between devices effectively prevent man-in-the-middle attacks and replay attacks, resolving the technical contradiction between convenience and security.

[0056] Specifically, in this example, two smart locks are installed on the main gate. Cameras are also installed along the path leading to the gate, capable of clearly capturing the faces of people passing by.

[0057] Both the smart lock and the camera are connected to and registered with the cloud platform, enabling network communication. The smart lock and camera are then set up as a bound pair. The platform records lock information including lock ID (e.g., A, B), lock IP, paired lock ID, paired lock IP, bound camera ID, and bound camera IP; camera information includes camera ID, camera IP, a list of bound lock IDs (including A and B), and a list of bound lock IPs. Both the lock and camera have their pairing and binding relationships recorded locally.

[0058] At a certain time, multiple people (single or multiple people) pass by the front camera of the house. The camera captures one or more faces, calculates facial feature vectors, and sends an "Immediate Visitor Notification Message" to two door locks. This message includes a message ID, message type (Immediate Visitor Notification Message), sender ID (camera), receiver ID (A and B respectively), a list of facial feature vectors (one or more), and a detection timestamp. Both door locks locally cache an "Immediate Visitor Information Table," containing the camera ID, facial feature vector list, and detection timestamp, based on a preset time period (e.g., 3 minutes, pre-set via the platform, estimated based on the normal walking time from the camera to the door with a certain margin). If the preset time period is exceeded, the door lock deletes the record from its cache.

[0059] When multiple people arrive at the gate, one person performs facial recognition authentication through any lock (e.g., A). Lock A obtains the person's facial feature vector and first compares it with a locally stored facial template. If a match is found, the person has the necessary permissions. Next, lock A compares the facial feature vector with each entry in the cached "Information Table of People About to Arrive" (all entries are within their validity period). If a match is found again, it means the person being authenticated has passed through the camera before arriving at the gate. At this point, lock A can open and simultaneously sends "facial recognition unlocking authentication information" to lock B, including message ID, message type (facial recognition unlocking authentication information), sender ID (A), receiver ID (B), the facial feature vector authenticated by lock A, and a message timestamp. After receiving the message, lock B also compares the facial feature vector with the locally stored facial template. If a match is found, lock B opens; if a match is not found, lock B sends a "facial unlocking authentication response message" to A, which includes the response message ID, message type, associated message ID, sender ID (B), receiver ID (A), response status (failed), and response message timestamp.

[0060] In the above process, if door lock A passes facial recognition but the face is not matched in the "Upcoming Visitor Information List," there are two possibilities: First, the person authenticating is authorized, but paused before reaching the door (e.g., answering a phone call), causing the facial recognition time to exceed the door lock's cache time for upcoming visitor information. Second, the person authenticating is a spoofing attacker; they did not spoof when passing the camera (the camera captured an unspoofed face), but only spoofed at the door lock, therefore the face obtained by door lock A is not in the "Upcoming Visitor Information List." To avoid the first scenario preventing authorized personnel from entering, door lock A, after passing facial recognition but not matching the face in the "Upcoming Visitor Information List," can provide a voice prompt: "The system has detected an anomaly; please continue fingerprint authentication!" (Note: Other authentication methods, such as password authentication or NFC cards, can also be used to achieve two-factor authentication for door lock A and reduce the risk of spoofing attacks.) Personnel authenticate their fingerprints using door lock A. Door lock A extracts the fingerprint features and compares them with locally stored fingerprint templates. If a match is found, door lock A can also be opened, and a "direct unlocking message" is sent to door lock B. This message includes a message ID, message type (direct unlocking message), sender ID (A), receiver ID (B), direct unlocking flag, and message timestamp. Upon receiving this message, door lock B can unlock.

[0061] In the above process, if lock A sends "face unlocking authentication information" to lock B after successful authentication, and lock B fails to detect the facial feature vector locally, then lock B sends "face unlocking authentication response information" to lock A (response status is failed). Lock A then provides a voice prompt: "The system has detected an abnormal situation. Please continue fingerprint authentication (or other methods)!" After that, lock A performs two-factor authentication. If the authentication is successful, lock A sends "direct unlocking information" to lock B for subsequent operations, which will not be elaborated further.

[0062] This application further proposes a cloud platform for recording door lock information, including a first smart lock ID, a second smart lock ID, a first smart lock IP, a second smart lock IP, a camera ID, and a camera IP; the camera records camera binding information, including a camera ID, a camera IP, a list of bound door lock IDs, and a list of bound door lock IPs, wherein the list of bound door lock IDs includes the first smart lock ID and the second smart lock ID, and the list of bound door lock IPs includes the first smart lock IP and the second smart lock IP.

[0063] The smart lock ID is a string encoding that uniquely identifies the smart lock device. It can be generated by encrypting the device serial number using a hash algorithm, and is used to distinguish different lock instances on the cloud platform. The smart lock IP address is the protocol address used by the smart lock in network communication. It can be implemented using IPv4 or IPv6 format, and is used to establish a direct communication link between devices. The bound lock ID list is a set of associated lock identifiers stored locally on the camera. It can be stored using a JSON array structure, and is used to restrict the camera to sending detection notifications only to specified locks. The bound lock IP list is the set of network addresses corresponding to the bound lock IDs. It can be implemented using a dynamic DNS resolution mechanism, and is used to ensure communication reachability between the camera and the lock.

[0064] Specifically, the cloud platform stores the mapping relationship between door lock IDs and IP addresses, providing the camera with a basis for device authentication and network addressing. When the camera detects a person, it filters the target device according to the locally bound door lock ID list, then establishes a communication connection through the IP list, and sends a first notification message to the first and second smart door locks. For example, when the camera ID is CAM001, its bound door lock ID list includes LOCK001 and LOCK002, with corresponding IP addresses of 192.168.1.101 and 192.168.1.102, respectively. Based on this list, the camera only sends data to these two IP addresses, avoiding the accidental sending of messages to other unbound door locks. The cloud platform synchronously updates device information; when the door lock IP changes, it automatically refreshes the binding list through a heartbeat mechanism to ensure the real-time effectiveness of the communication path.

[0065] Compared to existing technologies, traditional dual-lock linkage solutions rely on manual configuration of device relationship tables, which carries the risk of information update delays or configuration errors. This can lead to cameras sending messages to invalid IPs or missing target locks. This solution, however, uses a cloud platform to uniformly manage device identifiers and network addresses, combined with a dual verification mechanism based on the camera's local binding list, to achieve automated maintenance of device relationships and dynamic correction of communication paths.

[0066] Through the above technical solution, this application solves the problem of communication failure or false triggering caused by information inconsistency between the dual door lock and the camera, ensures that the detection notification is accurately delivered to the target door lock, reduces the unlocking error rate caused by IP change or equipment replacement, and reduces the operation cost of manually maintaining the device binding relationship.

[0067] This application further proposes a technical solution whereby, after receiving a first notification message from a camera, the first smart lock and the second smart lock cache the message for a preset time period, and delete the first notification message from the lock cache after the preset time period has elapsed. This includes:

[0068] In response to a person passing by the camera, the system captures a facial image based on the camera and determines a first facial feature vector, then sends a first notification message to the first smart lock and the second smart lock, including:

[0069] The first smart lock and the second smart lock cache the first notification message for a preset time period; if the preset time period is exceeded, the first smart lock and the second smart lock delete the first notification message from the lock cache.

[0070] The preset time period refers to the duration for which the door lock retains the first notification message. This can be implemented using a timer mechanism, such as setting it to 5 minutes or 10 minutes. This time period is set based on a reasonable time range for a person to move from the camera to the door lock, ensuring that valid data is retained within the recognition window. The door lock cache refers to the internal storage area of ​​the door lock used to temporarily store notification messages. This can be implemented using memory or flash memory modules, and the cache management module automatically triggers data cleanup operations.

[0071] Specifically, when the camera detects a person passing by and generates a first notification message, both the first and second smart locks receive the message and store it in their caches, simultaneously starting a timer for a preset period. If the person arrives at the lock within the preset time period, the lock will use the first notification message in the cache for facial feature comparison; if the person does not arrive after the timeout, the message in the cache is automatically deleted. This mechanism dynamically manages the lifecycle of cached data, ensuring that only data that can be used for identification within its validity period is retained, avoiding the waste of storage resources due to long-term storage of historical data. At the same time, deleting expired data reduces the computational load of subsequent feature matching, lowering the risk of misidentification due to invalid data participating in the comparison.

[0072] Compared to existing technologies, current solutions typically do not set a caching timeout for notification messages received by door locks, resulting in a large amount of expired data occupying storage space for extended periods and increasing unnecessary computational load. This solution introduces an automatic deletion mechanism triggered by a preset time period, accurately matching the time window of personnel movement, optimizing storage resource utilization while ensuring a high success rate of identification, and fundamentally reducing the possibility of misidentification.

[0073] Through the above technical solution, this application effectively solves the problem of storage resource waste caused by long-term caching of invalid notification messages in dual smart door locks. At the same time, by dynamically cleaning up expired data, it reduces the risk of misidentification in the feature matching process and improves the system's operating efficiency and recognition accuracy.

[0074] Furthermore, in response to a person arriving at the gate, a face detection is performed via a first smart door lock to obtain a second face feature vector. After verification by comparing the second face feature vector with a local face template, it is determined whether the second face feature vector exists in the first face feature vector list contained in the first notification message, and a first determination result is generated, including:

[0075] In response to a person arriving at the gate, the first smart door lock acquires the second facial feature vector and compares it with the locally stored facial template; if the comparison is successful, it determines whether the second facial feature vector exists in the first facial feature vector list contained in the first notification message.

[0076] The second facial feature vector refers to a mathematical vector generated by a feature extraction algorithm from a facial image captured by the camera of the first smart lock. Specifically, it can be implemented by using a convolutional neural network model to extract the coordinates of facial key points and converting them into a 128-dimensional floating-point array, used to represent the facial biometrics of the current user. The locally stored facial template refers to an authorized user feature dataset pre-stored in the lock's local memory. It can be stored as an encrypted binary file, with each template corresponding to the identity information of a registered user. The first notification message includes a list of first facial feature vectors, which is a set of all pedestrian feature data detected by the path camera within a preset time period. This data can be transmitted in JSON format via a wireless communication protocol, with each feature vector accompanied by a timestamp indicating the detection time.

[0077] Specifically, when a user approaches the gate, the camera on the first smart lock activates its real-time face capture function. Using edge computing, it encodes the captured image to generate a second facial feature vector. This vector is first compared to a locally stored encrypted face template library for similarity. If the match exceeds a preset threshold, primary authentication is triggered. Based on successful local verification, the system further retrieves the feature vector list from the received first notification message and uses vector space distance calculation to confirm whether the current user's features exist in the set of legitimate visitors pre-detected by the path camera. This dual verification mechanism retains the rapid response capability of the local database while establishing spatiotemporal correlation of access behavior through dynamic trajectory data.

[0078] Furthermore, regarding the processing of camera feeds as people pass by the gate: The camera records the facial feature vectors of all passing individuals and sends the information to both door locks (because if multiple people pass by simultaneously, it's unclear which person will authenticate at each lock or which lock will be used, so the facial feature vectors need to be recorded and sent to both locks). Both locks cache the facial feature vector information, resulting in wasted network bandwidth, lock cache resources, and power.

[0079] Compared to existing technologies, traditional dual-lock systems require users to perform a complete facial recognition process on each lock separately, significantly increasing operation time. This solution, however, sends the verification result to the secondary lock after the primary lock completes dual verification, allowing the secondary lock to perform only a simplified comparison process, effectively reducing user waiting time. Existing technologies that rely solely on path cameras or local lock databases for verification are vulnerable to forgery attacks or data tampering. This solution establishes a two-level cross-verification mechanism, requiring attackers to simultaneously bypass both the biometric data collection device and the network data transmission for unauthorized intrusion.

[0080] Through the above technical solution, this application enables legitimate users to unlock the secondary lock by performing facial recognition only once at the main lock, avoiding repeated verification operations. Simultaneously, through dual verification of dynamic trajectory verification and a local database, it effectively prevents attackers from breaching the access control system by forging biometric features or disrupting a single verification node, ensuring the overall security of the dual-lock system while improving ease of use.

[0081] Furthermore, in this embodiment, the method further includes:

[0082] In response to a person passing by the camera, the camera captures a facial image and calculates a first facial feature vector, generating a new record in a locally stored table of passing personnel information; wherein each record in the table of passing personnel information includes the first facial feature vector and a detection timestamp;

[0083] In response to a person arriving at the gate, the system performs face detection via the first smart lock, obtains a second face feature vector, calculates the estimated time point P after passing the camera, and sends a person confirmation request to the camera. The person confirmation request includes the face feature vector to be confirmed and the estimated time point P after passing the camera. The estimated time point P after passing the camera is the time taken for the first smart lock to perform face detection minus a preset duration. The preset duration is based on the time required to walk from the camera to the gate.

[0084] Based on the personnel confirmation request information, the camera searches the personnel information table for records that match the facial feature vector of the person to be confirmed; if a matching record exists, it verifies whether the detection timestamp of the record is within the preset range of the estimated time point P, and returns a response status to the first smart door lock; the response status includes hit and miss; wherein, if the verification passes, it is marked as hit, otherwise it is marked as miss;

[0085] The first smart lock unlocks according to the response status and sends a second notification message to the second smart lock; the second notification message includes a second facial feature vector.

[0086] The second smart door lock is controlled to compare the second facial feature vector with the local facial template, and the second smart door lock is unlocked according to the comparison result.

[0087] When a person passes by the camera, the camera captures a facial image and calculates a first facial feature vector, generating a new record in the locally stored table of passing personnel information. Each record contains the first facial feature vector and a detection timestamp. When the person arrives at the gate, the first smart lock performs facial detection to obtain a second facial feature vector, calculates the estimated time point P of passing the camera, and sends a personnel confirmation request to the camera. This request includes the facial feature vector to be confirmed and the estimated time point P. The camera retrieves matching records based on the request and verifies whether the detection timestamp is within the preset range of the estimated time point P, returning a hit or miss status to the first smart lock. The first smart lock unlocks based on the response status and sends a second notification message to the second smart lock. The second smart lock compares the second facial feature vector with a local template and then unlocks.

[0088] The estimated time point P, determined by the camera, refers to the time point calculated backwards from the current face detection time of the first smart door lock. This preset time can be set based on the average walking time from the camera to the door, for example, 30 seconds to 2 minutes. This time parameter is used to verify the spatiotemporal correlation of the person's movement trajectory. The person confirmation request information refers to structured data containing the biometric features to be verified and time constraints, which can be encapsulated in JSON or Protobuf format and used to trigger spatiotemporal consistency verification at the camera end. The preset range of the detection timestamp refers to a time window centered on the estimated time point P, for example, it can be set to 10 seconds before and after point P to tolerate walking time errors. This range is filtered for abnormal requests using a timestamp interval matching algorithm. The passed person information table refers to a structured data set stored locally on the camera, implemented using a hash table or relational database. Each record is associated with a facial feature vector and a detection time accurate to milliseconds, providing data support for spatiotemporal verification.

[0089] Specifically, when a person is captured by a camera along their path, a timestamped facial feature record is generated, forming a chain of evidence for their movement trajectory. Upon reaching the gate, the first lock dynamically generates a time window parameter based on a preset path duration and sends a time-constrained feature matching request to the camera. The camera uses a timestamp interval verification algorithm to verify if the requested features exist in the records within the specified time window. If the verification passes, the first lock unlocks and transmits the real-time facial features to the second lock. The second lock performs a secondary comparison using a local biometric template to ensure feature consistency. This process, through spatiotemporal constraints and multi-node collaborative verification, reduces user operation steps while constructing a defense mechanism against intrusion attacks.

[0090] Specifically, at a certain point in time, the camera captures one or more faces. After obtaining the facial feature vectors, it does not send messages to the two door locks, but instead stores a "Passing Person Information Table" locally, containing facial feature vectors and detection timestamps. Each successful detection of a person generates a new record. Records can be retained for a preset time (e.g., 30 minutes) and can be deleted after the timeout period.

[0091] When a person arrives at the gate, after the door lock A passes the face authentication, it sends a "person confirmation request information" to the camera, which includes message ID, message type (person confirmation request information), sender ID (A), receiver ID (camera), facial feature vector to be confirmed, and estimated time point when the person passes the camera (calculated by door lock A, i.e., the person's face authentication time point - preset time period, where the preset time period is the estimate based on the normal time required to walk from the camera to the gate, which is preset by the platform, for example, 3 minutes, and sent to the door lock for local storage). After receiving the message, the camera searches for all recorded facial feature vectors in its local "Passing Person Information Table" based on the facial feature vector in the message. If a match is found, it checks if the detection timestamp of the record falls within a certain time range before and after the estimated time point in the message (e.g., within 1 minute, configurable by the platform). If so, it indicates that the person in front of door lock A passed through the camera normally before arriving at door A. The camera then replies to door lock A with a response message containing the response message ID, associated message ID, sender ID (camera), receiver ID (A), response status (match), and response message timestamp. Upon receiving this message, door lock A opens. The subsequent process remains the same: A sends "facial recognition unlocking authentication information" to door lock B, which will not be elaborated further. If the camera misses the fingerprint scan, or if it does miss but the timestamp of the recorded data exceeds a certain time range before or after the estimated time, the camera sends a response message to door lock A with a "missed" status. Lock A then provides a voice prompt to the authentication personnel: "The system has detected an anomaly. Please continue fingerprint authentication (Note: or other authentication methods)!" and continue with two-factor authentication.

[0092] Compared to existing technologies, traditional dual-lock systems require users to complete two independent verification operations. This solution, however, establishes a spatiotemporal correlation verification mechanism between the camera and the dual locks, allowing legitimate users to trigger automatic collaborative verification of both locks with only a single biometric scan. While intruders in existing technologies may be able to gain entry by forcibly breaking into a single lock, this solution requires intruders to simultaneously bypass both the spatiotemporal verification at the camera and the biometric comparison of both locks, significantly increasing the difficulty of the attack.

[0093] Through the above technical solution, this application enables legitimate users to trigger automatic verification and opening of the dual locks during natural walking, eliminating the need for repeated operations. Simultaneously, intruders cannot forge facial feature data that conforms to spatiotemporal correlation, and must simultaneously bypass the independent biometric verification of both locks, effectively enhancing the level of security.

[0094] Furthermore, in some embodiments, if an attacker is posing as a legitimate person, they may do so only at the entrance (e.g., by wearing a mask befitting a legitimate person) and not while passing the camera. If a person starts wearing a mask as they pass the camera, it could lead to false positives. To mitigate this risk, the camera is optimized to not only identify and calculate the facial feature vector of the person passing by, but also to perform multi-dimensional verification of the person's static features (such as height) and dynamic features (such as walking posture) to determine whether they are a legitimate user (and a resident).

[0095] Data Training Phase: When a user passes by the camera, the camera not only calculates the facial feature vector from the captured image of the person, but also estimates the height using geometric relationships based on the camera's own height and angle, forming a static feature vector. Furthermore, it analyzes the person's gait through continuous frame analysis, using techniques such as skeletal tracking to extract joint positions and analyze features such as stride length, stride frequency, and arm swing amplitude, forming a dynamic feature vector (all of these are existing mature technologies and will not be detailed here). Initially, when a user passes by the camera, the camera locally records a "Passing Person Feature Information Table," containing the person's facial feature vector, static feature vector, dynamic feature vector, detection timestamp, and door lock verification status (initially empty). After both door locks are successfully authenticated and the door is opened, door lock A also sends a "Face Verification Passed Information" to the camera, containing a message ID, message type (Face Verification Passed Information), sender ID (A), receiver ID (camera), facial feature vector, authentication passed identifier, and message timestamp. Upon receiving this information, the camera marks the door lock verification status as passed in its local "Passing Person Feature Information Table" based on the facial feature vector. Subsequently, the person passed by the camera multiple times, and each time a new record for that person was added to the "Passing Person Feature Information Table," recording their facial feature vector, static feature vector, dynamic feature vector, detection timestamp, and door lock verification status (all passed). Through multiple recordings (approximately 5 times for static feature vectors and approximately 20 times for dynamic feature vectors; with subsequent new records, the static and dynamic feature vectors are continuously updated based on the most recent 5 and 20 records respectively), the camera statistical analysis obtains the average value of the static and dynamic feature vectors corresponding to specific facial feature vectors, and adds a new "Resident Person Feature Information Table" saved locally, containing facial feature vectors, the average static feature vector, the average dynamic feature vector, and the most recent passing time. This table verifies resident persons from multiple dimensions using facial feature vectors, static feature vectors, and dynamic feature vectors, preventing spoofing attacks that begin as soon as the camera passes by. Both the "Passing Person Characteristics Information Table" and the "Resident Characteristics Information Table" need to be dynamically updated. If the most recent passing time recorded in the "Resident Characteristics Information Table" exceeds a set value (such as 60 days), the camera will delete the record of that resident.

[0096] Based on the data training mechanism of the above cameras, the above scheme is further optimized as follows:

[0097] When a person arrives at the gate, door lock A, after facial recognition, sends a "Person Confirmation Request" to the camera. This request includes a message ID, message type (Person Confirmation Request), sender ID (A), receiver ID (camera), the facial feature vector to be confirmed, and the estimated time of passing the camera. Upon receiving this message, the camera, based on the facial feature vector and the estimated time of passing the camera, first searches its local "Person Passing Feature Information Table" for the facial feature vector within a certain time range before and after the estimated time (e.g., 1 minute before and after, configurable by the platform). If no match is found (possible scenarios: 1. A person posing as an attacker in front of the door lock but not posing when passing the camera; 2. A legitimate person who timed out after passing the camera and arriving at the gate), the camera replies to door lock A with a "Missed" status. Lock A then provides a voice prompt: "The system has detected an anomaly. Please continue fingerprint authentication (Note: or other authentication methods)!", and two-factor authentication continues.

[0098] If a match is found (indicating the person passed by the camera; possible scenarios: 1. a legitimate person and not a fake attacker; 2. a fake attacker), the camera continues to search the "Resident Feature Information Table" based on the facial feature vector (i.e., further determining if the person is a resident). If a match is found again (possible scenarios: 1. a resident and not a fake attacker; 2. a fake attacker), then within a certain time range before and after the estimated time point of passing by the camera (e.g., 1 minute before and after, configurable by the platform), consecutive frames are selected from the camera's cached video stream. Static and dynamic feature vectors are calculated separately and compared with the average static and dynamic feature vectors corresponding to the facial feature vector in the "Resident Feature Information Table." If the similarity is greater than a threshold (e.g., above 85%), it can be basically determined that the person is a resident and not a fake attacker. The camera replies with a response message to door lock A, indicating a match. After receiving this message, door lock A opens. The subsequent process is still A sending "facial unlocking authentication information" to door lock B, etc., which will not be elaborated further. If the search in the "Resident Feature Information Table" fails to find a match (possible situations include: 1. a legitimate person but not a resident; 2. a resident but in the data training phase, without generating a static feature vector mean or / and dynamic feature vector mean, therefore there is no record in the Resident Feature Information Table), or if the search in the "Resident Feature Information Table" finds a match but the static feature vector mean or / and dynamic feature vector mean similarity does not match (possible situations: 1. a resident but static or dynamic feature detection is incorrect; 2. spoofing attack), the camera will send a response message to door lock A with a "missed" status. Lock A will then issue a voice prompt: "The system has detected an anomaly. Please continue fingerprint authentication (Note: or other authentication methods)!" and continue with two-factor authentication.

[0099] Through the above methods, for permanent residents, the camera not only verifies their facial feature vectors, but also verifies static features (such as height) and dynamic features (such as walking posture), further effectively preventing spoofing attacks.

[0100] Furthermore, in some embodiments, after a person arrives at the gate and the door lock A performs facial authentication, it sends a "personnel confirmation request" to the camera. This improvement involves simultaneously coordinating with door lock B after door lock A performs facial authentication to verify other static features (height). Once verified, the door lock then sends the "personnel confirmation request" to the camera, further strengthening its own security verification.

[0101] Data Training Phase: Users perform facial authentication at any door lock (e.g., A). After door lock A calculates and verifies the facial feature vector, it analyzes the image captured by its camera to extract key points of the user's body (e.g., joints, head) to determine if the user is standing upright (a mature technology, not detailed here). If upright, door lock A combines its own camera height and the user's face to estimate height using geometric relationships, forming a static feature vector (a mature technology; height calculation is inaccurate for non-upright postures and is not performed). If A cannot determine the posture through image analysis, it can send a "Personnel Standing Posture Confirmation Request Message" to door lock B, containing the message ID, message type (Personnel Standing Posture Confirmation Request Message), sender ID (A), receiver ID (B), and message timestamp. After receiving the message, door lock B also performs image analysis through its camera to determine whether the person is standing upright (possible results: 1. Yes; 2. No; 3. Cannot be confirmed). It then sends a "Person Standing Upright Posture Confirmation Request Response Message" to door lock A, including the response message ID, associated message ID, sender ID (B), receiver ID (A), response status (Yes, No, Cannot be confirmed), static feature vector (calculated by B to form the person's static feature vector if the response status is Yes), and response message timestamp. If door lock A can calculate and obtain the static feature vector (height) of the verified person itself or obtain it from B's feedback, door lock A adds a record to its local "Successfully Authenticated Person Feature Information Table," including the face feature vector, static feature vector, and detection timestamp; simultaneously, it sends this information to door lock B, which also saves it locally (because subsequent users may also choose B for authentication, so data synchronization between the two door locks is necessary). Subsequently, regardless of whether the person passes facial recognition at A or B (if there is an anomaly, the door lock must implement a two-factor authentication strategy), and their posture is determined to be upright, their static feature vector (height) is calculated and a new record for that person is added to the local "Successfully Authenticated Personnel Feature Information Table," recording the facial feature vector, static feature vector, and detection timestamp. This record is also synchronously added to the other door lock. Through multiple recordings (e.g., 5 times), door lock A calculates the average value of the static feature vector corresponding to a specific facial feature vector and adds it to the "Resident Feature Information Table" on the door lock side, containing the facial feature vector, the average static feature vector, and the most recent authentication time. Both the "Successfully Authenticated Personnel Feature Information Table" and the "Resident Feature Information Table" need to be dynamically updated (e.g., children grow taller, requiring updates to their static feature vectors). If the most recent authentication time in the "Resident Feature Information Table" exceeds a set value (e.g., 60 days), the door lock deletes the resident's record.

[0102] Based on the data training mechanism of the two door locks mentioned above, the door lock face authentication process has been further optimized as follows:

[0103] When a person arrives at the gate, lock A (assuming the person chooses A for authentication) passes facial recognition. Similar to the data training phase described above, lock A uses its camera to analyze the image and determine if the person is standing upright. If so, it calculates a static feature vector (height). If A cannot determine this through image analysis, it can send a "Person Standing Posture Confirmation Request Message" to lock B. Upon receiving the message, lock B uses its camera to analyze the image and determine if the person is standing upright (possible results: 1. Yes; 2. No; 3. Cannot be confirmed), and sends a "Person Standing Posture Confirmation Request Response Message" back to A. If A can calculate and obtain the static feature vector (height) of the verified person itself or obtain it from B's feedback, it proceeds to the next step (otherwise, this process ends. It sends a "Passed Person Confirmation Request Information" to the camera and executes subsequent operations).

[0104] A retrieves the facial feature vector of the person from the local "Resident Feature Information Table". If a match is found (possible scenarios: 1. Legitimate person and not a spoofing attack; 2. Spoofing attack), the calculated static feature vector is compared with the average static feature vector in the table for similarity. If the similarity is greater than a threshold (e.g., above 85%), the person is considered legitimate and not a spoofing attack. A then sends a "Person Confirmation Request" to the camera and performs subsequent operations. If a match is found, but the static feature vector comparison is less than the threshold, it is considered a failure (possible scenarios: 1. The person is a resident, but the static or dynamic feature detection is incorrect; 2. Spoofing attack). A then provides a voice prompt: "The system has detected an anomaly. Please continue fingerprint authentication (Note: or other authentication methods)!" and continues with two-factor authentication. If A fails to find the facial feature vector of the person in the local "Residents Feature Information Table" (possible situations: 1. Resident, but still in the data training stage; 2. Non-resident, but a legitimate temporary user; 3. Impersonating an attacker), A will continue with two-factor authentication. After successful authentication, the record will be added to the "Successfully Authenticated Persons Feature Information Table".

[0105] Furthermore, in this embodiment, the gate also includes one or more third smart locks; the first smart lock is the main lock, equipped with a PIR sensor; the other locks are secondary locks; and both the main lock and the secondary locks store local face templates; the main lock and the secondary locks each use different face recognition models; the method also includes:

[0106] The master lock performs facial recognition on the user, generates a first recognition result, and randomly selects any secondary lock to send a facial recognition message carrying the first recognition result;

[0107] The secondary lock receives a face recognition message sent by the primary lock; the face recognition message contains a set of recognition results from both the primary and secondary locks.

[0108] If the number of successful recognitions in the recognition result set exceeds a preset threshold, then all smart locks are controlled to perform unlocking operations; otherwise, any other secondary lock that has not participated in the recognition is randomly selected to send a message carrying face recognition until the number of successful recognitions in the recognition result set exceeds the preset threshold or all locks have participated in the recognition.

[0109] The gate also includes one or more third smart locks. The first smart lock acts as the master lock and is equipped with a PIR sensor. The remaining locks act as secondary locks. Both the master and secondary locks store local face templates. After the master lock performs face recognition on the user and generates a first recognition result, it randomly selects any secondary lock to send a face recognition message carrying that result. The secondary locks receive the message and form a set containing the recognition results of both the master and secondary locks. When the number of successful recognitions in the set exceeds a preset threshold, all locks are unlocked. Otherwise, other secondary locks that have not yet participated in the recognition process are randomly selected for verification until the threshold condition is met or all locks have participated in the verification.

[0110] The PIR sensor, or passive infrared sensor, can be implemented using a pyroelectric infrared detector. It detects infrared radiation emitted by the human body to trigger the main lock to initiate the facial recognition process. The recognition result set is a set of verification data containing the facial comparison results between the main and secondary locks. This can be implemented using a distributed database or blockchain technology to dynamically count the verification results from multiple nodes. The preset threshold is the minimum number of successful verifications required to trigger the unlocking operation. It can be set to an integer value greater than or equal to 2 to balance security and operational efficiency. The random selection mechanism selects verification nodes from the secondary lock list using a pseudo-random algorithm. This can be implemented using a round-robin random number generator to prevent attackers from predicting the verification order.

[0111] Specifically, when the main lock detects a user approaching via its PIR sensor, it first performs local facial recognition to generate a preliminary verification result. Then, the system randomly selects any secondary lock that hasn't participated in the verification and sends the main lock's recognition result to that lock for secondary verification. After the secondary lock completes its local facial template comparison, it aggregates the verification results from both the main and secondary locks to form a recognition result set. When the number of successful matches in the set reaches a preset threshold, all locks simultaneously perform the unlocking operation. If the threshold is not reached, the system will randomly select other secondary locks for supplementary verification until the cumulative number of successful verifications meets the threshold requirement or all available secondary locks are traversed.

[0112] The above solution installs two locks on each main entrance. For more important buildings, more locks (e.g., five) can be installed, each using a different facial recognition model (trained with different algorithms and datasets). All five locks are connected to and registered with a cloud platform, enabling network communication. Each lock stores the IDs and IP addresses of the other four locks. One lock, designated as the master lock A (distinguished by its installation location), requires a large battery and a PIR (Passive Infrared Sensor) to activate facial recognition upon detecting an approaching person. The other four locks are positioned so that when a person stands in front of the master lock A for facial authentication, the other four locks can generally capture their face. The other four locks do not activate facial recognition by default; they only activate it after the other locks send a "facial recognition message" to conserve power.

[0113] By coordinating five door locks, a majority vote is held based on the authentication results of each lock to decide whether to unlock the door. If three locks pass authentication, the user is considered legitimate, and then an "unlock notification message" is sent to the next lock in turn to unlock it. This process continues until all locks are unlocked, providing a more secure way to detect masked spoofing attackers. Considering that relying on a single lock for voting would consume too much power, a relay-sending facial recognition message method is adopted. The specific solution is as follows:

[0114] (1) When a person performs face authentication in front of the main lock A, A obtains the person's facial feature vector and compares it with the locally stored face template. The recognition result is either yes or no. Lock A first randomly selects one of the other four locks (such as B) and sends a "face recognition message" to B, which includes message ID, message type (face recognition message), sender ID (A), receiver ID (B), A's recognition result (yes / no, i.e., whether A considers the person being authenticated to be a legitimate user), and message timestamp.

[0115] (2) After receiving the "face recognition message" sent by A, B starts face capture and generates a face feature vector. Using the face recognition model adopted by B, it compares the face with the face template stored locally, and the recognition result is yes or no (if the face feature vector cannot be successfully obtained, the recognition result is also no, the same below). B continues to select 1 from the other 3 door locks (such as C) and sends a "face recognition message" to C, including message ID, message type (face recognition message), sender ID (B), receiver ID (C), A's recognition result (yes / no), B's recognition result (yes / no), and message timestamp.

[0116] (3) Similarly, after receiving the “face recognition message” sent by B, C starts face shooting and generates face feature vector. Using the face recognition model adopted by C, it compares with the face template stored locally, and the recognition result is yes or no.

[0117] C combines the recognition results of A and B in the "face recognition message" sent by B with its own recognition result. If all the recognition results are positive, then more than half (3 out of 5 locks) have been recognized. C then unlocks the lock and continues to select one of the other 2 locks (such as D), and sends an "unlock notification message" to D, which includes message ID, message type (unlock notification message), sender ID (C), receiver ID (D), recognized lock IDs (A, B, C), unlocked ID (C), and message timestamp.

[0118] (4) D receives the "unlock notification message" sent by C. First, unlock the lock. Then, based on the identified lock ID and unlocked ID in the message, D determines that the object to which it will send the message is the last lock (E). Since the first unlocked ID in the message is not E (but C), D continues to send the "unlock notification message" to E, which includes the message ID, message type (unlock notification message), sender ID (D), receiver ID (E), identified lock IDs (A, B, C), unlocked IDs (C, D), and message timestamp.

[0119] (5) When E receives the “unlock notification message” sent by D, it first unlocks the lock. Then, based on the identified lock ID and unlocked ID in the message, E determines that the object to which it will send the message is A. However, the first unlocked ID in the message is not A (but C). Therefore, E continues to send the “unlock notification message” to A, which includes the message ID, message type (unlock notification message), sender ID (E), receiver ID (A), identified lock IDs (A, B, C), unlocked IDs (C, D, E), and message timestamp.

[0120] (6) Subsequently, similarly, when A receives the "unlock notification message", it first unlocks the lock, and then determines that the object to which it will send the message is B. However, the first unlocked ID in the message is not B (but C), so A continues to send the "unlock notification message" to B.

[0121] When B receives the "unlock notification message", it first unlocks the lock. Then it analyzes that the target of the message is C. Since the first unlocked ID in the message is C, B stops sending the "unlock notification message" (at this time, all locks are open).

[0122] (7) In step (3), C combines the recognition results of A and B in the "face recognition message" sent by B with its own recognition result. If one of the recognition results is negative or two are negative, then C continues to select one of the other two door locks (such as D) and sends a "face recognition message" to D, which includes message ID, message type (face recognition message), sender ID (C), receiver ID (D), recognition result of A (yes / no), recognition result of B (yes / no), recognition result of C (yes / no), and message timestamp.

[0123] (8) D receives the “face recognition message” sent by C, starts face shooting and generates face feature vector, uses the face recognition model adopted by D to compare with the face template saved locally, and the recognition result is yes or no.

[0124] Combining the recognition results of A, B, and C in the "face recognition message" sent by C, as well as its own recognition result, if the recognition result of 3 locks is negative, then the majority of recognitions have failed. D will stop unlocking and send a "face authentication failure message" to A, B, and C, containing the message ID, message type (face authentication failure message), sender ID (D), receiver ID (A / B / C), and message timestamp, notifying them to stop unlocking. If the recognition result of 2 locks is negative, D will continue to send a "face recognition message" to E, containing the message ID, message type (face recognition message), sender ID (D), receiver ID (E), recognition result of A (yes / no), recognition result of B (yes / no), recognition result of C (yes / no), recognition result of D (yes / no), and message timestamp. If the recognition result of 1 lock is negative (i.e., the recognition results of the other 3 locks are positive), D will send an "unlock notification message" to E. The subsequent process is as described above and will not be repeated.

[0125] (9) If E receives the "face recognition message" sent by D, it starts face capture and generates a face feature vector. Using the face recognition model adopted by E, it compares the face with the locally stored face template, and the recognition result is yes or no. Similarly, if the recognition result of 3 door locks is no, then the majority recognition fails, E stops unlocking and sends a "face authentication failure message" to A, B, C, and D to stop unlocking. If the recognition result of 2 door locks is no, then the majority recognition passes, E sends an "unlock notification message" to A, and the subsequent process is as described above, and will not be repeated.

[0126] If E receives an "unlock notification message" from D, it should be handled as described above.

[0127] Compared with the prior art, the traditional multi-door lock system requires users to complete the independent verification of each lock one by one. However, in this solution, the main lock actively triggers the verification process and introduces a cooperative verification mechanism for the secondary locks, enabling users to trigger the multi-lock linkage operation by only completing a single identification at the main lock. In the prior art, the risk that attackers can predict the verification path exists due to the fixed-order verification. This solution adopts a distributed verification method of randomly selecting secondary locks, effectively preventing attackers from concentrating on destroying specific verification nodes.

[0128] Through the above technical solution, this application realizes the operational convenience that legal users can trigger the multi-door lock linkage opening with a single verification, and at the same time ensures the validity of identity confirmation by at least two independent verification nodes through a dynamic threshold verification mechanism. In the face of brute force cracking or spoofing attacks, attackers need to break through multiple randomly selected verification nodes simultaneously to achieve illegal unlocking, significantly improving the overall anti-attack ability of the system. It has the following beneficial technical effects: 1. For legal personnel, they only need to perform face authentication on one door lock, and the other door lock can be automatically opened, simplifying the unlocking process and improving convenience. 2. Even if spoofing attackers deceive the verification of one door lock through spoofing attacks and other means, it is difficult for them to simultaneously meet the conditions of being recorded and verified successfully while passing normally in front of the camera (spoofing attackers generally do not know that the camera is involved in the verification process), improving the system security. 3. When the camera adopts face recognition and methods for recognizing height and walking posture, it further increases the difficulty for spoofing attackers and enhances security. 4. The door locks cooperate to judge whether the person stands upright. If so, the height is further estimated and compared locally, increasing the difficulty of spoofing attacks. 5. For more important houses, 5 door locks are installed, and different face recognition models are used. The decision of whether to unlock is made by voting in the way that the minority obeys the majority, which can better detect attackers with mask counterfeiting.

[0129] This application further proposes a secure unlocking system based on the linkage between a double-door lock and a camera, which is applied to a gate equipped with a first intelligent door lock and a second intelligent door lock, and a camera is installed on the access road to the gate. As Figure 2 shown, the system includes a registration module 201, a first notification module 202, a generation module 203, a second notification module 204, and an unlocking control module 205.

[0130] The system comprises several functional units: The registration module establishes communication relationships between devices, typically through a cloud platform's device registration interface. This involves submitting device identification codes for the door lock and camera to the cloud to establish network connections and binding relationships, forming a trusted device group. The first notification module pre-collects personnel features, using the camera's built-in facial recognition algorithm to extract feature vectors and pushing a timestamped feature list to the door lock's cache via a wireless communication protocol. The generation module verifies the identity of the main lock, using a feature vector matching algorithm to calculate the similarity between real-time detected facial features and a pre-stored list, generating a verification result. The second notification module triggers the secondary lock's verification, using message queue technology to encrypt and transmit the verified feature vectors to the secondary lock. The unlocking control module performs local verification on the secondary lock, using an embedded chip to compare feature vectors with stored templates and drive the lock mechanism based on threshold results.

[0131] Specifically, when an authorized user passes through the path camera, their facial features are extracted and pushed to the buffer of the dual-lock system. When the user reaches the main lock, the real-time detected facial features are matched against a pre-stored list for verification. If the verification is successful, the main lock opens and sends the feature data to the secondary lock. Upon receiving the feature data, the secondary lock independently performs a comparison verification against its locally stored template. If the verification is successful, the lock opens automatically. The entire process, through data linkage and step-by-step verification mechanisms between devices, automates the unlocking process, which originally required two independent operations, while maintaining the dual physical lock protection layers.

[0132] Compared to existing technologies, traditional dual-lock systems require users to perform two separate identity verification operations, which is cumbersome. This solution, however, utilizes a data linkage mechanism between pre-captured features from a camera and the locks to ensure independent verification security for both locks. It automatically triggers the secondary lock verification process after the primary lock's verification is successful, effectively reducing the number of user interactions. Existing dual-lock systems lack a mechanism for sharing verification information between devices, forcing users to repeatedly perform the same verification steps. This solution combines feature vector transmission with localized verification, maintaining the security of independent decision-making for each lock while eliminating the inconvenience of repetitive operations.

[0133] Through the above technical solution, this application enables the secondary lock to automatically complete verification and unlocking based on shared feature data after the user passes the primary lock verification, achieving simultaneous opening of both doors without additional operation. This solution maintains the physical protection strength of the dual locks while reducing the user's operation steps from two independent verifications to a single verification, effectively solving the problem of cumbersome operation for legitimate users. The system strikes a balance between security and convenience through data collaboration and step-by-step verification mechanisms between devices, preventing security vulnerabilities caused by simplified operation.

[0134] This application further proposes a cloud platform for recording door lock information, including a first smart lock ID, a second smart lock ID, a first smart lock IP, a second smart lock IP, a camera ID, and a camera IP; the camera records camera binding information, including a camera ID, a camera IP, a list of bound door lock IDs, and a list of bound door lock IPs, wherein the list of bound door lock IDs includes the first smart lock ID and the second smart lock ID, and the list of bound door lock IPs includes the first smart lock IP and the second smart lock IP.

[0135] The door lock information refers to the association data between smart door locks and cameras stored on the cloud platform. This can be implemented using a database table structure, where each record contains a unique device identifier and a network address for identity verification during device communication. The camera binding information refers to the list of authorized devices stored locally by the camera. This can be implemented using a key-value pair data structure, which, through a dual mapping of logical identifiers and physical addresses, restricts the camera to establishing communication links only with designated door locks.

[0136] This application further proposes a storage medium storing a computer program, which is loaded and executed by a processor to implement a secure unlocking method based on the linkage between dual locks and a camera. The storage medium refers to the physical carrier used to store the computer program code, specifically a solid-state drive, flash memory chip, or optical disc, serving to provide retrievable storage support for the linked unlocking process. The computer program refers to a set of code containing executable instructions, specifically written in a compiled binary file or scripting language, implementing the dual-lock collaborative verification logic through a preset algorithm. The processor loading and execution refers to the central processing unit parsing and running the program instructions, specifically using an embedded chip or a general-purpose computing chip, ensuring that the authentication, message passing, and unlocking control steps in the linked unlocking method are executed in a predetermined order.

[0137] Compared to existing technologies, traditional dual-lock systems require users to operate two locks separately for verification. This solution, however, uses program control to automatically trigger the secondary lock verification process after the primary lock is verified. Existing technologies involve independent operation of the two locks, leading to cumbersome procedures. This solution, through a programmatic collaborative mechanism, combines the two verification actions into a single user interaction while retaining the physical protection advantages of dual locks.

[0138] Through the above technical solution, this application enables the secondary lock to automatically perform verification and unlock after the main lock completes verification, avoiding repetitive operation steps. The security protection capability of the dual-lock system is maintained through programmatic collaborative verification, and the camera pre-identification and timestamp verification mechanism prevents unauthorized intruders from exploiting time differences for attacks. The program is embedded in the storage medium to ensure the reliable execution of the unlocking logic, and the processor loading process ensures the real-time performance and accuracy of the verification process.

[0139] This application further proposes a storage medium storing a computer program, which is loaded and executed by a processor to implement a secure unlocking method based on the linkage between a dual-door lock and a camera.

[0140] Storage medium refers to the physical carrier used to store computer program code, which can be implemented using solid-state drives, flash memory chips, or optical discs. Its function is to provide stable storage support for the linkage unlocking logic, ensuring that the program code can be repeatedly invoked. The computer program refers to a set of instructions containing the processes of camera facial feature acquisition, door lock feature comparison, and master / slave lock collaborative verification. It is specifically written in a programming language and compiled into an executable file. Its function is to transform the dual-door lock linkage control logic into operation instructions that can be recognized by the processor. Processor loading and execution refers to the process where the central processing unit reads the program code from the storage medium and runs it line by line. This can be implemented using an embedded processor or a general-purpose computing chip. Its function is to automate the linkage unlocking process through the program, replacing manual operation steps.

[0141] Through the above technical solution, this application enables automatic linkage unlocking of dual smart locks. Authorized users only need to perform a single authentication to open both locks, avoiding the inconvenience of repeated operations. Simultaneously, the two locks maintain independent authentication mechanisms; for example, the secondary lock requires local feature comparison to prevent unauthorized opening of the secondary lock after an attack on the primary lock, thereby maintaining the overall security of the system.

[0142] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A safe unlocking method based on the linkage of a double door lock and a camera, characterized in that, The application is applied to a gate installed with a first smart door lock and a second smart door lock; A camera is installed on a road leading to the gate; the method comprises: Connecting the first smart door lock, the second smart door lock, and the camera to a cloud platform network and registering with the cloud platform, and setting the first smart door lock, the second smart door lock, and the camera in a binding relationship; In response to a person passing through the camera, a first face feature vector is determined based on a face image captured by the camera, and a first notification message is sent to the first smart door lock and the second smart door lock; the first notification message contains a camera ID, a first face feature vector list, and a detection timestamp; In response to a person reaching the gate, a second face feature vector is obtained through face detection by the first smart door lock, and after verification by comparison with a local face template, it is determined whether the second face feature vector exists in the first face feature vector list contained in the first notification message, and a first determination result is generated; According to the first determination result, the first smart door lock is unlocked, and a second notification message is sent to the second smart door lock; the second notification message includes the second face feature vector; The second smart door lock is controlled to compare the second face feature vector with a local face template, and according to the comparison result, the second smart door lock is unlocked.

2. The method of claim 1, wherein the method further comprises: The cloud platform records door lock information; the door lock information includes a first smart door lock ID, a second smart door lock ID, a first smart door lock IP, a second smart door lock IP, a camera ID, and a camera IP; The camera records camera binding information; The camera binding information includes a camera ID, a camera IP, a binding door lock ID list, and a binding door lock IP list; the binding door lock ID list includes a first smart door lock ID and a second smart door lock ID; the binding door lock IP list includes a first smart door lock IP and a second smart door lock IP.

3. The method of claim 2, wherein the method further comprises: In response to a person passing through the camera, a first face feature vector is determined based on a face image captured by the camera, and a first notification message is sent to the first smart door lock and the second smart door lock, including: The first smart door lock and the second smart door lock are set to cache the first notification message for a preset time period; if the preset time period is exceeded, the first smart door lock and the second smart door lock delete the first notification message from the door lock cache.

4. The method of claim 3, wherein the method further comprises: In response to a person reaching the gate, a second face feature vector is obtained through face detection by the first smart door lock, and after verification by comparison with a local face template, it is determined whether the second face feature vector exists in the first face feature vector list contained in the first notification message, and a first determination result is generated, including: In response to a person reaching the gate, the first smart door lock obtains a second face feature vector and compares it with a locally saved face template; if the comparison is successful, it is determined whether the second face feature vector exists in the first face feature vector list contained in the first notification message.

5. The method of claim 1, wherein the method further comprises: The method further comprises: In response to the person passing through the camera, the camera captures a face image and calculates a first face feature vector, and generates a new record in a locally stored passing person information table; each record of the passing person information table contains a first face feature vector and a detection timestamp; In response to the person arriving at the gate, the first intelligent door lock performs face detection, obtains a second face feature vector, and calculates a passing camera estimated time point P, and sends a person confirmation request information to the camera; the person confirmation request information includes a to-be-confirmed face feature vector and a passing camera estimated time point P; the passing camera estimated time point P is the time of face detection of the current first intelligent door lock minus a preset time length; the preset time length is the time required to walk from the camera to the gate; The camera searches for a record matching the to-be-confirmed face feature vector in the passing person information table according to the person confirmation request information; if there is a matching record, it is verified whether the detection timestamp of the record is within the preset range of the estimated time point P, and a response state is returned to the first intelligent door lock; the response state includes hit and miss; if the verification is passed, it is marked as hit, otherwise it is marked as miss; The first intelligent door lock performs unlocking according to the response state, and sends a second notification message to the second intelligent door lock; the second notification message includes a second face feature vector; The second intelligent door lock is controlled to perform comparison between the second face feature vector and a local face template, and to perform second intelligent door lock unlocking according to the comparison result.

6. The method of claim 5, wherein the method further comprises: The gate further comprises one or more third intelligent door locks; the first intelligent door lock is a master lock and is configured with a PIR sensor; the remaining door locks are slave locks; and the master lock and the slave locks both store a local face template; The master lock and the slave lock both use different face recognition models; the method further comprises: The master lock performs face recognition on the user, generates a first recognition result, and randomly selects any slave lock to send a face recognition message carrying the first recognition result; The slave lock receives the face recognition message sent by the master lock; the face recognition message contains a recognition result set of the master lock and the slave lock; In response to the number of recognition passes in the recognition result set being greater than a preset threshold, all intelligent door locks are controlled to perform unlocking operation; otherwise, other any slave lock not participating in recognition is randomly selected to send a face recognition message, until the number of recognition passes in the recognition result set is greater than the preset threshold or all door locks have participated in recognition.

7. A safe unlocking system based on the linkage of a double door lock and a camera, characterized in that, Applied to a gate installed with a first intelligent door lock and a second intelligent door lock; A camera is installed on a road leading to the gate; the system comprises: A registration module for connecting the first intelligent door lock, the second intelligent door lock, and the camera to a cloud platform network, and registering with the cloud platform; and setting the first intelligent door lock, the second intelligent door lock, and the camera in a binding relationship; The first notification module is configured to, in response to a person passing through the camera, capture a face image based on the camera and determine a first face feature vector, and send a first notification message to the first smart door lock and the second smart door lock; the first notification message includes a camera ID, a first face feature vector list, and a detection timestamp; The generation module is configured to, in response to a person arriving at the gate, perform face detection through the first smart door lock, obtain a second face feature vector, and after passing the verification of comparison with a local face template, determine whether the second face feature vector exists in the first face feature vector list included in the first notification message, and generate a first determination result; The second notification module is configured to control the first smart door lock to be unlocked according to the first determination result, and send a second notification message to the second smart door lock; the second notification message includes the second face feature vector; The unlocking control module is configured to control the second smart door lock to perform comparison of the second face feature vector with the local face template, and perform unlocking of the second smart door lock according to a comparison result.

8. The dual mortise lock and camera linked security unlocking system of claim 7, wherein, The cloud platform records door lock information; the door lock information includes a first smart door lock ID, a second smart door lock ID, a first smart door lock IP, a second smart door lock IP, a camera ID, and a camera IP; The camera records camera binding information; The camera binding information includes a camera ID, a camera IP, a binding door lock ID list, and a binding door lock IP list; the binding door lock ID list includes a first smart door lock ID and a second smart door lock ID; and the binding door lock IP list includes a first smart door lock IP and a second smart door lock IP.

9. A storage medium storing a computer program; characterized by, The program is loaded and executed by the processor to implement the steps of the security unlocking method based on the linkage of the double door locks and the camera according to any one of claims 1-6.

Images