A community public space intelligent management method and device and a doorplate machine terminal

Abstract

The application discloses a kind of community public space intelligent management method, device and doorplate machine terminal, comprising: receiving and verifying the signature reservation token issued by cloud, and cache reservation record;According to the current semantic state determined by reservation time window, and determine reservation confidence score based on current semantic state;According to the preset advance of environmental equipment control parameter, send the control instruction of advance start to the environmental equipment in target space;In response to user arrival event, collect the biological characteristic image of user and process, obtain image quality score, live body confidence score and similarity score;According to current semantic state, reservation confidence score, image quality score, live body confidence score and historical behavior risk score, calculate to obtain dynamic identification threshold, and judge whether similarity score is greater than or equal to dynamic identification threshold, if yes, output unlocking instruction;According to the first time stamp of entry recorded by unlocking instruction, and when detecting that user leaves, record the second time stamp of departure, generate billing result.

Description

Technical Field

[0001] This invention relates to the field of smart community technology, and in particular to a smart management method, device and doorplate terminal for community public spaces. Background Technology

[0002] With the deepening development of smart communities, public spaces such as meeting rooms, gyms, and chess and card rooms have generally adopted a reservation-based management model, and have been equipped with facial recognition access control systems to achieve seamless access. Access control for community public spaces is characterized by a high degree of binding between access permissions and reservation records, exhibiting time-based, dynamic, and hierarchical characteristics. However, existing access control solutions generally treat facial recognition algorithms and business logic as independent modules, using fixed thresholds for recognition decisions. They cannot dynamically adjust the recognition threshold based on the user's reservation status (such as whether they are within the reservation period), leading to an irreconcilable conflict between security and user experience. Lowering the fixed threshold to ensure convenient access for reserved users reduces overall security; conversely, raising the threshold to improve security results in frequent rejection of legitimate reserved users. Furthermore, the control of environmental equipment (such as air conditioning and lighting) in existing community public spaces relies on passive sensors (such as PIR sensors), making it impossible to use reservation data for preheating or precooling. This results in a "cold start" delay for users upon arrival and an inability to adjust energy efficiency based on real-time usage, severely impacting the user experience. In terms of operational services, existing appointment payment models mostly adopt prepayment or per-use charging, which cannot accurately bill and reimburse based on actual passage behavior and duration, and also lack a linkage mechanism with the user credit system, resulting in high no-show rates and damaged operational revenue. Summary of the Invention

[0003] In view of this, the purpose of this invention is to provide an intelligent management method, device and doorplate terminal for community public spaces, in order to solve the above problems.

[0004] To achieve the above objectives, the present invention provides an intelligent management method for community public spaces, the method being applied to a doorplate terminal in the community public space, the method comprising: Receive and verify the signature reservation token issued by the cloud, and cache the reservation record carried by the signature reservation token, wherein the signature reservation token is generated by the cloud after the user completes the reservation, and the reservation record includes the user identity, reservation time window and environmental device control parameters; The current semantic state is determined based on the reservation time window, and the reservation confidence score is determined based on the current semantic state; According to the preset lead time of the environmental equipment control parameters, a control command for early start is sent to the environmental equipment in the target space, so that the target space switches from an idle standby state to a ready start state. The preset lead time is dynamically configured according to the type of environmental equipment. In response to a user arrival event, the system collects the user's biometric image, processes the biometric image, and obtains an image quality score, a liveness confidence score, and a similarity score. The dynamic recognition threshold is calculated based on the current semantic state, the reservation confidence score, the image quality score, the liveness confidence score, and the acquired user's historical behavior risk score. The similarity score is then determined to be greater than or equal to the dynamic recognition threshold. If so, an unlocking command is output and the target space is switched from the pre-startup state to the usage state. The first timestamp of the user's first entry is recorded according to the unlocking command, and the second timestamp of the user's last exit is recorded when the user leaves. A billing result is generated based on the first timestamp and the second timestamp.

[0005] To achieve the above objectives, the present invention also provides an intelligent management device for community public spaces, the device being applied to a doorplate terminal in the community public space, the device comprising: The reservation receiving unit is used to receive and verify the signature reservation token issued by the cloud, and cache the reservation record carried by the signature reservation token. The signature reservation token is generated by the cloud after the user completes the reservation, and the reservation record includes the user identity, reservation time window and environmental device control parameters. A semantic parsing unit is used to determine the current semantic state based on the reservation time window, and to determine the reservation confidence score based on the current semantic state; An environmental pre-control unit is used to send a pre-start control command to the environmental equipment in the target space according to the preset lead time of the environmental equipment control parameters, so that the target space switches from an idle standby state to a ready-to-start state, wherein the preset lead time is dynamically configured according to the type of environmental equipment. The biometric processing unit is used to collect the user's biometric image in response to the user arrival event, process the biometric image, and obtain image quality score, liveness confidence score and similarity score; The dynamic decision unit is used to calculate a dynamic recognition threshold based on the current semantic state, the reservation confidence score, the image quality score, the liveness confidence score, and the acquired user's historical behavior risk score, and to determine whether the similarity score is greater than or equal to the dynamic recognition threshold. If so, it outputs an unlocking command and switches the target space from the pre-startup state to the usage state. The billing management unit is used to record the first timestamp of the user's first entry based on the unlocking command, and to record the second timestamp of the user's last exit when the user leaves the premises, and to generate a billing result based on the first timestamp and the second timestamp.

[0006] To achieve the above objectives, the present invention also proposes a doorplate terminal, which includes a processor, a memory, and a computer program stored in the memory. The computer program is executed by the processor to implement the steps of an intelligent management method for community public spaces as described in the above embodiments.

[0007] To achieve the above objectives, the present invention also proposes a computer-readable storage medium storing a computer program that is executed by a processor to implement the steps of an intelligent management method for community public spaces as described in the above embodiments.

[0008] Beneficial effects: The above solutions ensure the integrity and tamper-proof nature of reservation data by receiving and verifying signed reservation tokens, dynamically integrating reservation semantics into access control decisions, and realizing the transformation from fixed rule control to semantic-driven control. By activating environmental equipment in advance based on preset lead times for environmental equipment control parameters, cold start delays are eliminated, improving user experience. Dynamic recognition thresholds are calculated by combining image quality scores, liveness confidence scores, and historical behavior risk scores, achieving synergistic optimization of security and user experience. Precise billing based on access timestamps forms a complete business loop for access control and operational services, thereby comprehensively improving the security level, ease of use, and refined level of space operation in community public spaces.

[0009] By acquiring multiple frames with quality perception and fusion, and selecting high-quality and effective frames for weighted fusion, the image usability in harsh scenarios such as low light, motion, and occlusion is improved. The fusion of visible light and near-infrared dual features enhances the robustness of recognition under different lighting conditions. The dual-branch detection of physical attacks and digital forgery attacks by the shared encoder improves the comprehensiveness of liveness detection. By using multiple frames with quality perception and fusion and local sensitive hash transformation, irreversible encrypted storage of biometric features is achieved while ensuring recognition accuracy. This not only improves the robustness of recognition in complex scenarios but also meets privacy protection compliance requirements, achieving a balance between security and convenience.

[0010] By employing a multi-dimensional dynamic threshold calculation formula, factors such as reservation confidence, liveness confidence, image quality, and historical behavioral risks are quantified into specific adjustment parameters, breaking the limitations of traditional access control systems that rely on a single fixed threshold. By setting tiered base thresholds for confirmed reservation states, near-reservation states, and non-reservation states, the system can adaptively adjust the access difficulty based on the urgency and security requirements of the scenario. This ensures convenient access for legitimate users while effectively blocking unauthorized access, thereby maximizing the access experience for reserved users while maintaining high security during non-reservation periods. In particular, a linear decay mechanism is introduced during the grace period after the reservation time window ends, avoiding the problem of legitimate users being mistakenly blocked or security vulnerabilities caused by hard switching. This allows the recognition threshold to transition smoothly over time, providing users with a buffer time to organize their belongings after the reservation ends and preventing the risk of unauthorized intrusion due to excessively low thresholds after the reservation expires, thus achieving flexible adjustment of the security strategy.

[0011] By providing alternative access paths in case of facial recognition failure—that is, in the confirmed reservation state, when facial comparison fails, a dynamic QR code backup access is automatically triggered—highly reliable entry for reserved users is ensured. This effectively addresses facial recognition failure scenarios caused by temporary obstruction or extreme lighting, improving the system's fault tolerance and success rate, and protecting the legitimate rights of reserved users from being affected by malfunctions. In the usage state, energy-saving control (reducing lighting, increasing air conditioning temperature) is achieved through personnel presence detection during temporary vacancy periods. After the reservation ends and the personnel leave, the system switches to a closed or deep sleep state, enabling dynamic allocation of spatial environmental resources on demand. Compared to traditional passive sensing solutions, this improves energy efficiency, minimizing energy costs in public spaces while ensuring user comfort.

[0012] Using the actual passage timestamp as the billing anchor, the system accurately calculates the actual usage time, distinguishes between normal billing and overtime billing, and avoids fee disputes under the traditional prepayment model. A no-show determination mechanism is established (no passage record and no prior cancellation), automatically calculating late fees and deducting credit points. User levels are dynamically adjusted through credit points, granting different levels differentiated reservation priorities, cancellation grace periods, and fee deduction rights, forming an effective incentive and constraint mechanism that effectively reduces no-show rates, improves space resource utilization, and ultimately increases operational revenue.

[0013] By deeply integrating reservation service semantics with intercom call routing, the system dynamically binds the list of responsible administrators based on reservation records for the current time period and initiates video calls after checking online status in order of priority. By querying the list of responsible administrators in the reservation records in real time and combining it with online status detection, the system ensures that call requests can accurately reach the management personnel currently responsible for the space, improving the response speed and connection rate of property services. At the same time, when all administrators are offline or unresponsive, the system automatically escalates to a broadcast call and triggers an emergency notification, solving the problem that traditional building intercom fixed extension numbers cannot be routed in a contextualized manner. Through a multi-level escalation call and emergency notification mechanism, the system provides reliable emergency communication support for community public spaces and fills the gap in intelligent intercom functionality for conference room doorplate machines. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a flowchart illustrating an intelligent management method for community public spaces according to an embodiment of the present invention.

[0016] Figure 2 This is a schematic diagram of the structure of an intelligent management device for community public spaces provided in an embodiment of the present invention.

[0017] The realization of the invention's objective, its functional characteristics, and advantages will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to represent selected embodiments of the invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] The present invention will be described in detail below with reference to the embodiments.

[0020] Reference Figure 1 The diagram shown is a flowchart illustrating an intelligent management method for community public spaces according to an embodiment of the present invention. This method is applied to doorplate terminals in community public spaces; specifically, the method includes: S11, Receive and verify the signature reservation token issued by the cloud, and cache the reservation record carried by the signature reservation token, wherein the signature reservation token is generated by the cloud after the user completes the reservation, and the reservation record includes the user identity, reservation time window and environmental device control parameters.

[0021] The signed reservation token is HMAC signed using a device-specific key, and multiple signed reservation tokens form a hash chain through the hash value of the previous token. The signed reservation token includes a reservation identifier, a user identifier, a reservation time window, environmental device control parameters, and responsible administrator information.

[0022] In this embodiment, after a user completes the reservation of a community public space and deducts the prepayment through a mini-program or a third-party reservation system, the cloud generates a Signed Reservation Token (SRT) for the reservation. The payload of the token encapsulates the reservation identifier, user identifier, reservation time window (including start and end time), preset advance time for various types of environmental equipment (e.g., air conditioning starts pre-cooling 30 minutes in advance, heating starts pre-heating 45 minutes in advance, lighting gradually brightens 5 minutes in advance, projector turns on 3 minutes in advance, etc.), and the person in charge or administrator of the space during this time period (e.g., administrator account, SIP number, mobile phone number, and a list of administrators sorted by priority). To ensure the integrity and tamper-proof nature of reservation data from the cloud to the doorplate terminal, the cloud uses a device-specific key to perform HMAC-SHA256 signature on the token payload and links the current token with the SHA-256 hash value of the previous reservation token, forming a hash chain. This means that each token's payload stores the hash value of the previous token. Any tampering with historical reservation records will break the hash chain, allowing the system to detect anomalies and trigger alarms. Upon receiving the signed reservation token from the cloud, the doorplate terminal first verifies the signature's validity using the device-specific key stored in its trusted execution environment and checks the continuity of the hash chain. After successful verification, it caches the parsed reservation records locally (typically for the next 24 hours), extracts environmental device control parameters for subsequent environmental pre-control, and extracts a list of responsible administrators for intelligent intercom call routing. In addition, the terminal continuously detects abnormal patterns in the cached reservation records, including high-frequency reservations (such as the same user making multiple reservations for different spaces within a short period of time), time overlaps (the same user making multiple reservations for spaces within the same time period), inconsistent user-face bindings, and cross-device replay attacks. Once an anomaly is detected, the user's current semantic state is immediately forcibly switched to a non-reservation state, and the information is reported to the cloud for review. Through these steps, the system achieves reliable transmission and local caching of reservation data, providing a reliable data foundation for subsequent functions such as dynamic threshold decision-making based on reservation semantics, proactive environment preparation, and accurate billing.

[0023] S12, determine the current semantic state according to the reservation time window, and determine the reservation confidence score based on the current semantic state.

[0024] S13, according to the preset lead time of the environmental equipment control parameters, a control command for early start is sent to the environmental equipment in the target space, so that the target space switches from idle standby state to ready start state, wherein the preset lead time is dynamically configured according to the type of environmental equipment.

[0025] In this embodiment, the doorplate terminal determines the semantic state corresponding to the current moment based on the reservation time window in the locally cached reservation records. Specifically, the semantic state is divided into three levels: confirmed reservation state, near-reservation state, and no-reservation state. The confirmed reservation state corresponds to the time interval from the first preset duration (e.g., 5 minutes) before the reservation start time to the second preset duration (e.g., 2 minutes) after the reservation end time; the near-reservation state corresponds to the time interval from the third preset duration (e.g., 30 minutes) before the reservation start time to the first preset duration (e.g., 5 minutes); and the no-reservation state is the time outside of the above two intervals. Based on the determined semantic state, a reservation confidence score is further determined: if the current state is confirmed, the reservation confidence score is between 0.8 and 1.0 (specifically, it can be linearly mapped according to the proportion of remaining reservation time); if the current state is near-reservation, the reservation confidence score is between 0.4 and 0.8; and if the current state is no-reservation, the reservation confidence score is 0. This reservation confidence score will serve as one of the important input parameters for subsequent dynamic identification threshold calculation, reflecting the degree of matching between the current user and the reservation record.

[0026] By parsing the reservation records in the cache, the system extracts the environmental equipment control parameters contained therein. These parameters are configured with preset lead times for different types of environmental equipment, such as the pre-cooling lead time for the air conditioning system (starting 30 minutes in advance in cooling mode), the pre-heating lead time for the heating system (starting 45 minutes in advance), the activation lead time for the lighting system (linearly increasing from 5% brightness to the preset brightness 5 minutes in advance in a gradual brightening mode), the activation lead time for the projector or display screen (starting 3 minutes in advance to eliminate cold start delay), and the activation lead time for the fresh air system (switching to low ventilation 20 minutes in advance). Based on the start time of the current reservation time window, the system subtracts the preset lead time for the corresponding equipment to calculate the absolute start time of each equipment, and sends an advance start control command to the corresponding environmental equipment in the target space when the time point arrives. The control commands are converted into a device-recognizable format through a built-in hardware protocol adaptation layer (supporting KNX, BACnet / IP, Modbus RTU / TCP, Zigbee 3.0, Z-Wave, and cloud-to-cloud integration protocols such as Mijia, Tuya, and Huawei HiLink), enabling precise control of devices such as air conditioners, lighting, projectors, and fresh air systems. After executing the above control commands, the target space switches from an idle standby state to a ready-to-start state.

[0027] Idle standby mode refers to a state where all devices are in deep energy-saving mode (e.g., lights are off, air conditioning is set to 25°C and maintained, and projectors are in deep sleep mode) when there are no current reservations and no reservations within the next 2 hours. Ready-to-start mode means the system starts the relevant devices in advance according to the reservation schedule, gradually bringing environmental parameters such as temperature and brightness in the space closer to a comfortable level, thus eliminating the 3-15 minute cold start delay caused by starting the devices after people arrive. If a temporary reservation is inserted during the pre-cooling or pre-heating process (e.g., a reservation originally scheduled for 2:00, the system started pre-cooling at 1:30, but an emergency meeting from 1:30 to 2:00 is inserted at 1:25), the system does not cancel the pre-cooling already executed, but directly reuses the existing pre-cooling results for the emergency meeting and keeps the air conditioning running until both meetings are finished, avoiding a secondary cold start caused by switching in between. If pre-cooling has not yet started, it immediately starts in advance and prepares as much as possible with the remaining time, while simultaneously sending a notification message to the user.

[0028] Furthermore, the system supports collaborative energy-saving scheduling for multiple adjacent public spaces: when the reservation times of two adjacent spaces overlap, reservations are prioritized for spaces with lower heat loads to reduce simultaneous full-power operation; for multiple spaces sharing air conditioning or fresh air units, the optimal equipment start-up and shutdown plan is calculated based on their respective reservation times to avoid frequent start-ups and shutdowns that could damage equipment lifespan; corridor lighting automatically turns on to guide brightness (maintaining 60% brightness for 10 minutes before the reservation starts and 5 minutes after it ends) based on the expected entry and exit times of the reserved personnel, and maintains 20% duty brightness for the remaining time periods, which can save 15% to 25% energy compared to traditional passive sensing solutions. Through the above steps, a paradigm shift from passive sensing and response to proactive prediction and preparation is achieved, improving the user experience and enhancing overall energy efficiency.

[0029] S14, in response to the user arrival event, collect the user's biometric image, process the biometric image, and obtain image quality score, liveness confidence score and similarity score.

[0030] Furthermore, in step S14, the biometric image is processed to obtain an image quality score, a liveness confidence score, and a similarity score, including: S14-1, using the formula The image quality score of a single frame is calculated, where S is the image quality score of the frame. sharpness S represents the resolution score. illumination S represents illumination fraction. pose S represents the deflection angle minute. occlusion The occlusion score is represented by θ1, θ2, θ3, and θ4, which represent the corresponding weighting coefficients. S14-2, Select valid frames with image quality scores greater than or equal to a preset threshold, perform weighted fusion to obtain a fused image, extract visible light features and near-infrared features from the fused image, input the visible light features and near-infrared features into the physical attack detection branch and digital forgery detection branch of the shared encoder for fusion, and output the liveness confidence score; S14-3, extract the facial feature vector of the fused image, map the facial feature vector into an irreversible binary hash code through local sensitive hash transformation, and search and match the binary hash code in a preset feature library to obtain the similarity score.

[0031] Updating the output of the liveness confidence score includes: Using consecutive frames of biometric images as a window, the temporal sequence of shallow convolutional features of visible light and near-infrared features in consecutive frames of biometric images is extracted; Calculate the cross-modal cross-correlation coefficient in the temporal sequence of the shallow convolutional features to obtain the temporal consistency score; Based on facial landmark detection, the temporal sequence of eye opening and closing in multiple consecutive frames of biometric images is calculated. The authenticity of micro-expressions in the temporal sequence of eye opening and closing is evaluated by a lightweight LSTM network to obtain a micro-expression authenticity score. The temporal consistency score and the micro-expression authenticity score are fused into the liveness confidence score to obtain the updated liveness confidence score.

[0032] In this embodiment, when a user approaches the doorplate terminal in the community's public space, the terminal's built-in proximity sensor (such as PIR or frame difference method) detects the approaching person and triggers the image acquisition process. The terminal simultaneously activates both the RGB visible light camera and the near-infrared (NIR) camera, continuously acquiring five frames of RGB images and their corresponding NIR images within a 300-millisecond time window. For each frame, its single-frame image quality score C is first evaluated. quality The formula is The specific calculation methods for each sub-indicator are as follows: S sharpness Sharpness score is calculated by applying Laplacian filtering to the image, calculating the variance of the pixel values ​​after filtering, truncating the variance to the range of [0, 500], and then normalizing by dividing by 500 to obtain a sharpness score between 0 and 1. The closer the variance value is to 500 (the richer the image details), the closer the score is to 1. When the variance is 0, the score is 0.

[0033] S illuminationIllumination score: The acquired image is converted to grayscale, and the average brightness of all pixels in the image is calculated. The comfortable brightness range is set to [100, 180] (0~255 grayscale levels). If the average is between 100 and 180, the score is 1.0. If the average is below 100, the score decreases linearly: score = average / 100. If the average is above 180, the score decreases linearly: score = 1 - (average - 180) / (255 - 180), and the minimum is 0. This process ensures that images that are too dark (average < 100) and overexposed (average > 180) are penalized.

[0034] S pose The yaw and pitch angles of the face are obtained through facial landmark detection, and each angle is truncated to the interval [-45°, 45°]. A score is then calculated for each angle. angle =1 - |angle| / 45; The average of the scores for yaw and pitch angles is the final S. pose For example, a frontal view (0°, 0°) scores 1; a deflection of ±45° scores 0; and a deflection exceeding ±45° is calculated as ±45°.

[0035] S occlusion Occlusion score representation: The face region is divided into five areas using the face location confidence map (PCM): eyebrow and eye area, bridge of nose area, lower face area, ear area, and forehead area. A lightweight neural network is used to regress the visibility confidence score (range 0-1, 0 indicating complete occlusion, 1 indicating complete visibility). The arithmetic mean of the confidence scores of the five areas is calculated as S. occlusion If the confidence score of a certain region (such as the eyebrow area) is lower than 0.3, the system will automatically switch to the recognition mode dominated by near-infrared images.

[0036] After the above four sub-indicators are calculated, they are multiplied by the corresponding weight coefficients (such as θ1=0.30, θ2=0.30, θ3=0.20, θ4=0.20) and summed to obtain the final single-frame image quality score, which ranges from 0 to 1.

[0037] For each image frame, if its image quality score is lower than a preset threshold (e.g., 0.6), the frame is discarded. If the number of valid frames is insufficient within a 300-millisecond window, the system triggers a supplementary light for re-acquisition; if the number is still insufficient, it is downgraded to the QR code alternative access channel. Valid frames (quality score ≥ 0.6) are weighted and fused according to their respective quality score proportions to generate a fused image. If the current ambient illumination is lower than a preset threshold, the system first performs histogram equalization and frequency domain brightness compensation (processing delay not exceeding 15 milliseconds) before proceeding with subsequent processing. Visible light features and near-infrared features are extracted from the fused image and simultaneously input into a dual-branch network of a shared encoder: a physical attack detection branch detects physical attack traces such as printed photos, screen moiré patterns, and paper masks; a digital forgery detection branch detects frequency domain artifacts and pixel inconsistencies in faces generated by generative adversarial networks or diffusion models. The outputs of the two branches are dynamically weighted according to the modal signal quality and then fused to obtain a preliminary liveness confidence score. To further enhance anti-spoofing capabilities, the system also integrates remote photoplethysmography (PPG) analysis: when the current semantic state is non-appointed or the preliminary liveness confidence score is within the boundary range of 0.65 to 0.75, chroma time-series signals are extracted from the face region (forehead and bilateral cheekbone regions of interest) of the fused image. A chroma space-based PPG extraction algorithm is used to remove motion noise and perform a Fast Fourier Transform to extract the peak heart rate. If the detected heart rate is within the effective range of 60 to 120 beats per minute, a positive correction coefficient (e.g., an increase of 0.10) is applied to the preliminary liveness confidence score.

[0038] Simultaneously, the system performs multimodal temporal consistency liveness verification: using 30 consecutive frames (approximately 1 second) as a window, shallow convolutional features of visible light flow and near-infrared flow are extracted to form a temporal sequence. Cross-modal cross-correlation vectors with delays of 0 to 3 frames are calculated, and Mahalanobis distance is calculated with a statistical model of the actual face distribution. If the Mahalanobis distance exceeds 2.5, a temporal anomaly is determined. Furthermore, based on facial landmark detection, the temporal signal of eye opening and closing is calculated. A lightweight long short-term memory network with fewer than 200K parameters and a delay of less than 20 milliseconds is used to evaluate the authenticity of micro-expressions. If there are no micro-expression changes (such as blinking or squinting) for more than 5 consecutive seconds, a low authenticity score is output. The temporal consistency score and micro-expression authenticity score are fused into the preliminary liveness confidence score to obtain the final liveness confidence score.

[0039] While performing liveness detection, the system extracts facial feature vectors from the fused image. This process employs a lightweight student model (MobileNetV3-Face, approximately 4.2 million parameters, with a model size of approximately 1.8 megabytes after INT8 quantization) deployed on an edge neural network processing unit. This student model is trained from the teacher model (ArcFace-ViT-B, approximately 86 million parameters) through progressive knowledge distillation. During distillation, a semantically aware distillation loss function is used, assigning higher distillation weights to pre-arranged samples and known attack samples (pre-arranged samples have a weight of 1.5, and attack samples have a weight of 2.0), ensuring that the student model maintains a high degree of consistency with the teacher model in core scenarios. The student model's inference latency is approximately 180 milliseconds.

[0040] The extracted 512-dimensional original facial feature vector is then transformed using a privacy-preserving method: In a trusted execution environment, a user-specific random matrix jointly generated by the device identifier and user identifier is used. Locality-sensitive hashing (LSH) is then applied to map the feature vector to a 512-bit binary hash code, where hash code = sign(random matrix × feature vector). This hash code is irreversible, and when revocation is needed, simply replacing the random matrix renders the original template invalid. The system retrieves this binary hash code from a hash code library that is also encrypted with AES-256-CBC, performs fast matching using Hamming distance, and returns Top-K candidate records along with their corresponding similarity scores (similarity scores are based on Hamming distance transformation and range from 0 to 1).

[0041] S15, calculate the dynamic recognition threshold based on the current semantic state, the reservation confidence score, the image quality score, the liveness confidence score, and the acquired user's historical behavior risk score, and determine whether the similarity score is greater than or equal to the dynamic recognition threshold. If so, output the unlocking command and switch the target space from the pre-start state to the usage state.

[0042] Further, in step S15, the calculation of the dynamic identification threshold based on the current semantic state, the reservation confidence score, the image quality score, the liveness confidence score, and the acquired user's historical behavior risk score includes: S15-1, using formula Calculations are performed to obtain an initial dynamic threshold. This initial dynamic threshold is then limited to a preset numerical range to obtain the dynamic recognition threshold; where T... base P represents the basic threshold determined by the current semantic state. booking E represents the reservation confidence score. liveness For the liveness confidence score, C quality R is the image quality score. historyFor the historical behavior risk score, α, β, γ, and δ represent the corresponding weight coefficients; wherein, the current semantic state includes confirmed appointment state, near-appointment state, and no appointment state; when the current semantic state is confirmed appointment state, the basic threshold is a first preset value; when the current semantic state is near-appointment state, the basic threshold is a second preset value; when the current semantic state is no appointment state, the basic threshold is a third preset value; the first preset value is less than the second preset value, and the second preset value is less than the third preset value.

[0043] The confirmed reservation state corresponds to the time interval from the first preset time before the reservation starts to the second preset time after the reservation ends; the near reservation state corresponds to the time interval from the third preset time before the reservation starts to the first preset time; and the non-reservation state is the time interval other than the confirmed reservation state and the near reservation state.

[0044] Furthermore, when the current semantic state is a confirmed reservation state, it also includes: Within a preset grace period after the end of the appointment time window, the formula is used. Calculations are performed to obtain the updated initial dynamic threshold, where T is... S1 T represents the first preset value. S3 This represents the third preset value, t. end t represents the end time of the reservation time window. grace This indicates the preset grace period, where t represents the current time, and t end <t≤t end +t grace ; The updated initial dynamic threshold is limited to the preset numerical range to obtain the dynamic recognition threshold.

[0045] Furthermore, in step S15, determining whether the similarity score is greater than or equal to the dynamic recognition threshold includes: S15-2, when the similarity score is less than the dynamic recognition threshold and the current semantic state is a confirmed reservation state, the alternative access process is triggered; wherein, triggering the alternative access process includes displaying a dynamic QR code bound to the current reservation on the doorplate terminal, scanning and verifying the dynamic QR code within a preset validity period, and outputting an unlocking command after successful verification.

[0046] Furthermore, in step S15, after switching the target space from the pre-startup state to the usage state, the process further includes: S15-3, when the target space is in use and no personnel are detected within a preset time period, the target space is switched from the use state to the temporary vacancy state, and an energy-saving control command is sent to the environmental equipment, wherein the energy-saving control command includes adjusting the lighting brightness to a preset brightness and adjusting the air conditioning temperature to a preset temperature. S15-4, When the reservation time window ends and personnel are detected leaving the site, the target space is switched from the usage state to the usage end state, and the environmental equipment is turned off or switched to deep energy-saving mode.

[0047] In this embodiment, the basic threshold T is determined based on the current semantic state. base The current semantic state is divided into three levels: confirmed appointment, near-appointment, and no appointment. The confirmed appointment corresponds to the time interval from the first preset duration (e.g., 5 minutes) before the appointment start time to the second preset duration (e.g., 2 minutes) after the appointment end time. The near-appointment appointment corresponds to the time interval from the third preset duration (e.g., 30 minutes) before the appointment start time to the first preset duration (e.g., 5 minutes). The no-appointment state is the time outside of the above two intervals. For the confirmed appointment, the base threshold is the first preset value (e.g., 0.60); for the near-appointment, the base threshold is the second preset value (e.g., 0.68); and for the no-appointment, the base threshold is the third preset value (e.g., 0.80). The first preset value is less than the second preset value, and the second preset value is less than the third preset value.

[0048] Obtain the calculated image quality score, liveness confidence score s, similarity score, and user historical behavior risk score R retrieved from the cloud or local cache. history (This score combines the number of recognition failures, temporal anomalies, and short-term high-frequency attempt anomalies of the user over the past 30 days, and normalizes them to the 0-1 range.) Then, the initial dynamic threshold is calculated using the following dynamic threshold formula: , where P booking The reservation confidence score determined in step S12 based on the semantic state is 0.8~1.0 for confirmed reservations, 0.4~0.8 for near-reservation reservations, and 0 for non-reservation reservations; in some implementations, P bookingIt can also be obtained by mapping a multi-dimensional semantic state vector containing user role level, space authorization level, time risk coefficient, and device trust level through a configurable policy matrix, so as to perform fine-grained security relaxation or tightening for different users / spaces in the same state; α, β, γ, and δ are preset weight coefficients, which respectively control the degree to which the reservation confidence score lowers the threshold (e.g., α ranges from 0.08 to 0.15), the degree to which the liveness confidence score tightens the threshold (e.g., β ranges from 0.05 to 0.10), the degree to which the image quality score tightens the threshold (e.g., γ ranges from 0.03 to 0.08), and the degree to which the historical behavior risk score raises the threshold (e.g., δ ranges from 0.05 to 0.15). Each coefficient can be determined by offline validation set grid search to ensure that the false rejection rate in the confirmed reservation state meets the design target.

[0049] For cases where the current semantic state is a confirmed reservation, the system further introduces a decay mechanism: a preset grace period t after the reservation time window ends. grace Within a certain timeframe (e.g., 2 minutes), the initial dynamic threshold is updated using a linear decay formula: when t... end <t≤t end +t grace When, T_dynamic_initial(t) = T S1 + (T S3 - T S1 ) × (t - t end ) / t grace In the formula, T S1 T represents the first preset value (basic threshold for confirmed appointment status). S3 This represents the third preset value (the basic threshold for the non-reserved state), t. end t represents the end time of the appointment window. grace This represents the preset grace period, where t represents the current time. This attenuation causes the threshold to decrease from T during the grace period. S1 Smooth transition to T S3 This avoids the security risks or disrupted user experience caused by a hard switch after the reservation ends. After the attenuation is complete, the system will process the updated initial dynamic threshold and the initial dynamic threshold obtained from the aforementioned formula in a unified manner.

[0050] After obtaining the initial dynamic threshold (whether calculated normally or updated with decay), it is restricted to a preset numerical range (e.g., a lower limit of 0.55 and an upper limit of 0.92), i.e., T. dynamic = clip(T_dynamic_initial, 0.55, 0.92), finally obtaining the dynamic recognition threshold. After completing the dynamic threshold calculation, perform a similarity score comparison: if the similarity score S match Greater than or equal to the dynamic recognition threshold T dynamicIf the identification is successful, an unlocking command is output, and the target space's state is switched from the pre-start state to the usage state. At the moment of unlocking, the access control controller drives the electric lock to open the space access control.

[0051] If the similarity score S match Less than the dynamic recognition threshold T dynamic If the current semantic state is a confirmed reservation, the alternative access process is triggered. At this time, the doorplate terminal screen automatically displays a dynamic QR code bound to the current reservation. The QR code contains a reservation hash value generated by the signed reservation token (to prevent forgery), and its validity period is, for example, 60 seconds. The user scans the QR code using a mobile app, and the system verifies its validity before outputting an unlocking command. If the similarity score is less than the threshold and the current semantic state is an adjacent reservation or non-reservation state, access is denied and a failure log is recorded. Simultaneously, the user's historical behavior risk score R is updated. history (Increase the number of failures, count short-term high-frequency attempts, etc.).

[0052] After a user successfully unlocks and enters the space, the target space enters the usage state. The system continuously monitors whether anyone is present in the space using the human detection capability described in step S14. If no one is detected in the usage state for a first preset time period (e.g., 10 minutes), the target space is switched from the usage state to a temporary vacancy state. At this time, an energy-saving control command is sent to the environmental equipment: the lighting brightness is reduced to 30%, the air conditioning temperature setting is increased by 2 degrees Celsius (increased by 2°C for cooling scenarios, decreased by 2°C for heating scenarios), and a "Space Unoccupied Warning" is pushed to the user's mobile app, prompting the user whether they need to extend the reservation or release the space. When the system detects that someone has re-entered, it automatically switches back to the usage state.

[0053] When the appointment window closes and the system detects that the person has left (through exit detection logic, such as no face or body detected for 30 consecutive seconds), the target space is switched from the usage state to the usage-finished state. In this state, the lighting is turned off, the air conditioning is switched to energy-saving standby mode (e.g., maintaining 25℃), and a "to be cleaned" notification is pushed to the cleaning staff's app. If the next appointment start time for the current space is more than 2 hours away, a deep energy-saving mode is executed, and all devices are adjusted to the lowest power consumption standby. In addition, the system also supports collaborative energy-saving scheduling for adjacent spaces: when the appointment time periods of two adjacent spaces overlap, the appointment is prioritized for the space with lower heat load to avoid peak load caused by both spaces entering full power mode at the same time; for multiple spaces sharing the same air conditioning or fresh air unit, the system calculates the optimal equipment start-stop plan based on the appointment time sequence of each space to reduce frequent equipment start-stop; the corridor lighting automatically turns on the guiding brightness according to the expected entry and exit time of the appointment personnel (maintaining 60% brightness from 10 minutes before the appointment starts to 5 minutes after the appointment ends), and maintains 20% duty brightness for the rest of the time, saving 15%~25% energy compared to the traditional PIR sensing solution.

[0054] S16, record the first timestamp of the user's first entry according to the unlocking command, and record the second timestamp of the user's last exit when the user leaves, and generate a billing result based on the first timestamp and the second timestamp.

[0055] Furthermore, in step S16, generating the billing result based on the first timestamp and the second timestamp includes: S16-1, Calculate the actual usage time using the first timestamp as the billing start time and the second timestamp as the billing end time; S16-2, If the actual usage time does not exceed the reservation time window, the fee is calculated at the normal unit price; otherwise, the overtime fee is calculated at the overtime unit price, and the user's credit score is updated based on the billing result. S16-3 If the user does not generate a passage record within the reservation time window and does not cancel the reservation in advance, it is determined as a no-show, and a penalty is calculated and the corresponding points are deducted from the credit score; wherein, the credit score is dynamically adjusted according to the no-show, lateness, on-time performance, and early cancellation, and user levels are divided according to the credit score. Different user levels have different reservation priorities, cancellation exemption periods and fee deduction rights.

[0056] In this embodiment, when a user successfully verifies their identity via facial recognition or a QR code (optional method) and triggers the unlocking command, the doorplate terminal records the current system time as the first timestamp of the user's initial entry, which serves as the billing start time. Subsequently, a continuous presence detection mechanism (e.g., a lightweight human detection every 30 seconds) monitors the space for any presence. If no facial or human features are detected for a continuous period (e.g., 30 seconds), it is determined that the user has left, and the current time is recorded as the second timestamp of the last exit, which serves as the billing end time. Using the first timestamp as the billing start time and the second timestamp as the billing end time, the actual usage time (accurate to the second) is calculated. Then, the user's reservation time window (start and end times) is retrieved from their reservation record, and the actual usage time is compared with the reservation time window.

[0057] If the actual usage time does not exceed the reserved time window, the normal unit price will be used for calculation. For example, if the normal unit price for a space is 2 yuan / hour, and the user actually uses it for 45 minutes, the cost will be calculated as 0.75 hours, which is 1.5 yuan. If the actual usage time exceeds the reserved time window, the excess time will be charged at the overtime unit price, which is usually 1.5 times the normal unit price (e.g., 3 yuan / hour). The total cost will be obtained by adding the normal time period cost and the overtime time period cost. Regardless of whether the billing is normal or overtime, the corresponding amount will be deducted from the user's linked payment account based on the final cost, and the user's credit score will be updated. If the user fulfills the reservation on time and there is no overtime, the credit score will increase (e.g., +2 points); if there is an overtime but it does not exceed a certain limit, the credit score will remain unchanged or decrease slightly (specific rules can be configured).

[0058] Meanwhile, the system also detects no-shows: if a user does not generate any access records within the reservation time window (i.e., no first timestamp or any successful identification record), and the user does not actively cancel the reservation in advance (e.g., more than 2 hours before the reservation starts), it is considered a complete no-show. The system calculates a penalty of 30% to 50% of the reservation amount (e.g., 40% of the total reservation fee), deducted from the user's pre-authorization or credit account, and deducts a corresponding amount from their credit score (e.g., 10 points are deducted for each complete no-show). If a user is more than 30 minutes late and does not apply for a delay in advance, it is considered a late no-show, and a partial penalty is calculated based on the proportion of lateness to the reservation duration, and credit points are deducted (maximum 5 points). If a user leaves early within the reservation time slot for more than 30% of the reservation duration, the remaining time slot is automatically released for others to reserve, and a refund is issued for the user who leaves early (e.g., 60% of the remaining time slot fee is refunded, and 40% is retained as platform operating costs). At the same time, credit score adjustments are made at the system's discretion based on the performance of the user.

[0059] The system maintains a credit score for each user. The dynamic adjustment rules for the credit score include: deductions: complete no-show (e.g., -10 points / time), late no-show (-2 to -5 points / time), facility damage record (-20 points / time), short-term high-frequency failed identification attempts (-3 points / time, determined by historical behavior risk score R). history (Triggered by linkage); Bonus points: Arriving and leaving on time (+2 points / time), proactively canceling reservations in advance (cancellation more than 2 hours before the reservation starts, +1 point / time), participating in facility satisfaction rating (+0.5 points / time), no no-show records for 3 consecutive months (+5 points / quarter). Based on credit scores, users can be divided into different levels: VIP users with credit scores of 130 or above enjoy priority reservation rights (priority confirmation when there is competition for reservations in the same time slot), a grace period for late cancellation (cancellation within 30 minutes before the reservation starts without penalty, 2 hours before for regular users), and points to offset reservation fees (100 points can offset 1 free use), etc.; users with credit scores below 60 are blacklisted and reservation privileges can only be restored after manual review.

[0060] For operation platforms managing multiple public spaces, the system also supports flexible pricing: independently configure billing prices based on time periods (peak hours on weekdays, off-peak hours, weekends) and space types (meeting rooms, gyms, card rooms). When space utilization is below 30%, a discount price is automatically triggered to attract reservations. It also supports unified settlement for subscription packages (monthly cards, quarterly cards) and corporate accounts, with package fees deducted from the credit account and unused credit carried forward monthly (up to 2 months). All billing results, credit score changes, no-show records, etc., are aggregated in the cloud to generate an operational data dashboard (number of reservations, actual utilization rate, revenue statistics, no-show rate trends, high-frequency user rankings, etc.), and can automatically synchronize financial data through API integration with the property management ERP system.

[0061] In another embodiment, the method further includes: S17, in response to the intercom call request triggered by the user on the doorplate terminal, query the list of responsible administrators bound in the reservation record for the current time period, the list of responsible administrators including at least one administrator account sorted by priority; S18, detect the online status of each administrator in the list of responsible administrators, and initiate a video call to any online administrator based on the online status; S19. If all administrators in the current time period's list of responsible administrators are offline or unresponsive, initiate a broadcast call and trigger an emergency notification to be sent to the preset management personnel.

[0062] The system detects the online status of each administrator in the list of responsible administrators, including querying through a cloud-based heartbeat detection interface. If there is no response within the response timeout threshold, the administrator is determined to be offline.

[0063] In this embodiment, the doorplate terminal integrates an intelligent intercom calling function, specifically including: when a user presses the "Call Operations Management Personnel" button on the doorplate terminal (the button display name can be customized in the Web backend according to the venue type, such as "Call Clubhouse Manager" or "Contact Property Management," and supports hidden configuration), the system responds to the intercom call request by first querying the list of responsible administrators bound to the reservation records for the current time period. This list is dynamically bound by the cloud platform according to the space configuration and time period information when the reservation is generated, and is encapsulated in the payload of the signed reservation token (admin_list field), containing at least one administrator account sorted by priority (supporting multiple forms such as Liling management machine account, App account, SIP number, mobile phone number, etc.). If there are multiple shift administrators in the reservation records for the current time period, the list can contain a maximum of 3 administrators sorted by priority.

[0064] The system then sequentially checks the online status of each administrator in the list of responsible administrators: It queries the real-time online status of each administrator account (e.g., whether the management machine is online, whether the mobile app is active in the foreground) through the heartbeat detection interface provided by the cloud platform, and sets a response timeout threshold, such as 2 seconds. If no online response is received from the administrator within the timeout threshold, they are considered offline. The system checks administrators in order of priority, from highest to lowest: First, it checks the top (preferred) administrator in the list. If they are online, a SIP / WebRTC video call is immediately initiated to them. If they are offline or do not respond, the process is automatically escalated to the next-ranked administrator in the list, and the process is repeated. If the next-ranked administrator is also offline, the system continues to check the third-ranked administrator. During the video call, the system supports full-duplex audio and video communication, with a minimum resolution of 720P, a frame rate of 25fps, and audio using Opus codec, with an end-to-end latency of no more than 300 milliseconds. Administrators can answer calls on their mobile app or management machine, view the live feed from the doorplate camera in real time, and remotely perform one-click door opening operations (sending an opening command to the doorplate terminal) through the management interface, without needing to be physically present.

[0065] If all administrators in the current time period's list of responsible administrators (including backups) are determined to be offline or unresponsive (e.g., each administrator waits 15 seconds without responding), an automatic escalation broadcast call will be initiated, simultaneously calling all online administrators (not limited to those in the list), and displaying "Contacting management personnel, please wait" on the doorplate screen. If there is still no response after 60 seconds following the broadcast call, an emergency notification will be triggered, sending alarm information (including location, time, and call records) via SMS or app to a pre-set emergency contact (such as the property duty manager), while simultaneously displaying the property service phone number on the screen for users to dial themselves.

[0066] In addition, the system also supports an emergency scenario priority mechanism: when a user presses the call button more than 3 times in a row with an interval of less than 10 seconds between each press, it is automatically identified as an emergency call scenario and the call priority is immediately raised to the highest level. This means that all online status detection steps are skipped and a broadcast call is initiated directly. At the same time, an App notification (with a real-time screenshot of the doorplate machine) is pushed to all administrators to ensure a rapid response in emergency situations.

[0067] The system comprehensively logs the entire process of each intercom call, including: call initiation time, routing decision process (which administrators were checked in sequence, the online status of each administrator, and the administrator account that was actually connected), call start and end times, call duration, and whether an emergency SMS was triggered. These logs are periodically reported to the cloud and used by the property management to generate KPIs such as "call response rate," supporting service quality analysis and refined operation management.

[0068] The method further includes: The image quality distribution, matching similarity distribution, liveness detection distribution, and inference latency of the doorplate terminal are statistically analyzed in real time and weighted to obtain the device health score; When the device health score is lower than a first threshold, local calibration is automatically performed, wherein the local calibration includes adjusting image preprocessing parameters or renormalizing feature distribution; when the device health score is lower than a second threshold, a cloud alarm or on-site maintenance is triggered.

[0069] In this embodiment, the system also integrates a device-level online adaptive calibration mechanism to ensure the stability of the identification performance of the doorplate terminal throughout its entire lifecycle. Specifically, it calculates health sub-scores from four dimensions by real-time statistical analysis of the doorplate terminal's operating data over the past 7 days, and then performs a weighted summation to obtain the device health score H_device. The first dimension is image quality health H_quality (weight, e.g., 0.30), calculated by statistically analyzing the average quality score μ_current of all face image frames collected in the past 7 days (the quality score is calculated in the same way as C in step S14). quality And compare it with the baseline quality mean μ_baseline calibrated at the time the equipment left the factory; H_quality = 100 × max(0, 1 -|μ_current - μ_baseline| / μ_baseline). For example, if the current average quality score is 20% lower than the baseline, then H_quality is approximately 60 points.

[0070] The second dimension is the health of the similarity distribution, H_match (with a weight of 0.35). This is calculated by statistically analyzing the mean μ_match and standard deviation σ_match of the similarity scores obtained from all recognition requests over the past 7 days, forming a normal distribution. This distribution is then compared to the factory baseline distribution (also with mean μ_base_match and standard deviation σ_base_match) to calculate the KL divergence, KL_match: H_match = 100 × exp(-KL_match / 0.3). A larger KL divergence indicates a more severe drift in recognition and matching performance, resulting in a lower score.

[0071] The third dimension is the liveness detection health score H_liveness (weight 0.20). It is calculated by statistically analyzing the distribution of the liveness confidence score E_liveness of each identification output in the past 7 days, and also by calculating the KL divergence KL_liveness with the baseline distribution. H_liveness = 100 × exp(-KL_liveness / 0.3).

[0072] The fourth dimension is inference latency health, H_latency (weighted at 0.15). This is calculated by statistically analyzing the end-to-end latency from image acquisition to output for each recognition session over the past 7 days, taking the P90 quantile (i.e., 90% of requests have latency below this value), and comparing it to the design target latency of 500 milliseconds: H_latency = 100 × min(1, 500 / P90_latency). If the P90 latency is 500 milliseconds, the score is 100; if it is 1000 milliseconds, the score is 50.

[0073] The four sub-scores are weighted and summed as follows: H_device = 0.30×H_quality + 0.35×H_match +0.20×H_liveness + 0.15×H_latency; H_device ranges from 0 to 100, with higher scores indicating healthier devices.

[0074] Three health thresholds are preset: when H_device is below 85, a local lightweight calibration is triggered. This local calibration does not modify the neural network backbone weights but employs two unsupervised methods: first, adjusting image preprocessing parameters, such as optimizing the Gamma correction value, to minimize the KL divergence between the brightness distribution of the currently acquired image and the baseline distribution; second, renormalizing the extracted feature vectors using the formula f_calibrated = (f - μ_device) / σ_device × σ_base + μ_base, where μ_device and σ_device are the mean and standard deviation of the device's features over a recent period, and μ_base and σ_base are the factory baselines. After calibration, the calibration effect is self-supervised and verified: the KL divergence of the image brightness distribution or feature distribution before and after calibration is compared. If it decreases, calibration is accepted and device parameters are updated; otherwise, the state before calibration is rolled back. When H_device is below 70, the health score and anomaly logs are reported to the cloud, where cloud maintenance personnel remotely analyze and decide whether a firmware update or patch needs to be pushed. When H_device scores below 55, a field maintenance alarm is triggered. Simultaneously, a message indicating that the device requires maintenance is displayed on the doorplate screen, and the recognition policy is forcibly switched to strict mode (e.g., increasing the overall dynamic threshold by 0.05 and disabling the QR code alternative channel) to maximize security. Through this mechanism, the system can promptly detect and automatically correct device performance drift, reducing the manual maintenance costs of large-scale deployments and ensuring long-term stable operation.

[0075] Reference Figure 2 The diagram shown is a structural schematic of an intelligent management device for community public spaces provided in an embodiment of the present invention.

[0076] In this embodiment, the device is applied to a doorplate terminal in a community public space. The device 20 includes: The reservation receiving unit 21 is used to receive and verify the signature reservation token issued by the cloud, and cache the reservation record carried by the signature reservation token. The signature reservation token is generated by the cloud after the user completes the reservation, and the reservation record includes the user identity, reservation time window and environmental device control parameters. The semantic parsing unit 22 is used to determine the current semantic state based on the reservation time window, and to determine the reservation confidence score based on the current semantic state; The environmental pre-control unit 23 is used to send a control command for early start to the environmental equipment in the target space according to the preset advance amount of the environmental equipment control parameters, so that the target space switches from an idle standby state to a ready start state, wherein the preset advance amount is dynamically configured according to the type of environmental equipment; The biometric processing unit 24 is used to collect the user's biometric image in response to the user arrival event, process the biometric image, and obtain image quality score, liveness confidence score and similarity score; The dynamic decision unit 25 is used to calculate a dynamic recognition threshold based on the current semantic state, the reservation confidence score, the image quality score, the liveness confidence score and the acquired user's historical behavior risk score, and to determine whether the similarity score is greater than or equal to the dynamic recognition threshold. If so, it outputs an unlocking command and switches the target space from the pre-start state to the usage state. The billing management unit 26 is used to record the first timestamp of the user's first entry based on the unlocking command, and to record the second timestamp of the user's last exit when the user leaves the premises, and to generate a billing result based on the first timestamp and the second timestamp.

[0077] Each unit module of the device 20 can execute the corresponding steps in the above method embodiment, so the details of each unit module will not be elaborated here. Please refer to the description of the corresponding steps above for details.

[0078] This invention also provides a door number terminal, which includes a processor, a memory, and a computer program stored in the memory. The memory stores software programs and modules, and the processor executes various functional applications and data processing by running the software programs and modules stored in the memory. The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the use of the device, etc. Furthermore, the memory may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other volatile solid-state storage device. Accordingly, the memory may also include a memory controller to provide access to the memory for the processor and input unit.

[0079] The input unit can be used to receive input numerical, character, or image information, and to generate keyboard, mouse, joystick, optical, or trackball signal inputs related to user settings and function control. Specifically, in addition to a camera, the input unit of this embodiment may also include a touch-sensitive surface (e.g., a touch screen) and other input devices.

[0080] The display unit can be used to display information input by the user or information provided to the user, as well as various graphical user interfaces of the device. These graphical user interfaces can be composed of graphics, text, icons, video, and any combination thereof. The display unit may include a display panel, optionally configured as an LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), or other similar display panel. Furthermore, a touch-sensitive surface may cover the display panel. When the touch-sensitive surface detects a touch operation on or near it, it transmits the information to the processor to determine the type of touch event. Subsequently, the processor provides corresponding visual output on the display panel based on the type of touch event.

[0081] This invention also provides a computer-readable storage medium, which may be a computer-readable storage medium included in the memory described in the above embodiments; or it may be a standalone computer-readable storage medium not assembled into a device. The computer-readable storage medium stores at least one instruction, which is loaded and executed by a processor to implement... Figure 1 The diagram illustrates an intelligent management method for community public spaces. The computer-readable storage medium may be a read-only memory, a hard disk, or an optical disk, etc.

[0082] This invention also provides a computer program product, including a computer program / instructions, which are loaded and executed by a processor to implement... Figure 1 This illustrates an intelligent management method for community public spaces.

[0083] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the device embodiments, equipment embodiments, and storage medium embodiments, since they are basically similar to the method embodiments, the descriptions are relatively simple, and relevant parts can be referred to the descriptions in the method embodiments.

[0084] Furthermore, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0085] The foregoing description illustrates and describes preferred embodiments of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the inventive concept by means of the foregoing teachings or techniques or knowledge in related fields. Any modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A method for intelligent management of community public spaces, characterized in that, The method is applied to a doorplate terminal in a community public space, and the method includes: Receive and verify the signature reservation token issued by the cloud, and cache the reservation record carried by the signature reservation token, wherein the signature reservation token is generated by the cloud after the user completes the reservation, and the reservation record includes the user identity, reservation time window and environmental device control parameters; The current semantic state is determined based on the reservation time window, and the reservation confidence score is determined based on the current semantic state; According to the preset lead time of the environmental equipment control parameters, a control command for early start is sent to the environmental equipment in the target space, so that the target space switches from an idle standby state to a ready start state. The preset lead time is dynamically configured according to the type of environmental equipment. In response to a user arrival event, the system collects the user's biometric image, processes the biometric image, and obtains an image quality score, a liveness confidence score, and a similarity score. The dynamic recognition threshold is calculated based on the current semantic state, the reservation confidence score, the image quality score, the liveness confidence score, and the acquired user's historical behavior risk score. The similarity score is then determined to be greater than or equal to the dynamic recognition threshold. If so, an unlocking command is output and the target space is switched from the pre-startup state to the usage state. The first timestamp of the user's first entry is recorded according to the unlocking command, and the second timestamp of the user's last exit is recorded when the user leaves. A billing result is generated based on the first timestamp and the second timestamp.

2. The intelligent management method for community public spaces according to claim 1, characterized in that, The process of processing the biometric image to obtain image quality score, liveness confidence score, and similarity score includes: Using formula The image quality score of a single frame is calculated, where S is the image quality score of the frame. sharpness S represents the resolution score. illumination S represents illumination fraction. pose S represents the deflection angle minute. occlusion The occlusion score is represented by θ1, θ2, θ3, and θ4, which represent the corresponding weighting coefficients. Valid frames with image quality scores greater than or equal to a preset threshold are selected and weighted fused to obtain a fused image. Visible light features and near-infrared features of the fused image are extracted. The visible light features and near-infrared features are input into the physical attack detection branch and digital forgery detection branch of the shared encoder for fusion, and the liveness confidence score is output. The facial feature vector of the fused image is extracted, and the facial feature vector is mapped into an irreversible binary hash code through Local Sensitive Hash Transform. The binary hash code is then searched and matched in a preset feature library to obtain the similarity score.

3. The intelligent management method for community public spaces according to claim 1, characterized in that, The dynamic recognition threshold is calculated based on the current semantic state, the reservation confidence score, the image quality score, the liveness confidence score, and the acquired user's historical behavior risk score, including: Using formula Calculations are performed to obtain an initial dynamic threshold. This initial dynamic threshold is then limited to a preset numerical range to obtain the dynamic recognition threshold; where T... base P represents the basic threshold determined by the current semantic state. booking E represents the reservation confidence score. liveness For the liveness confidence score, C quality R is the image quality score. history For the historical behavior risk score, α, β, γ, and δ represent the corresponding weighting coefficients; The current semantic state includes a confirmed appointment state, an approaching appointment state, and a non-appointment state. When the current semantic state is a confirmed appointment state, the basic threshold is a first preset value; when the current semantic state is an approaching appointment state, the basic threshold is a second preset value; when the current semantic state is a non-appointment state, the basic threshold is a third preset value; the first preset value is less than the second preset value, and the second preset value is less than the third preset value.

4. The intelligent management method for community public spaces according to claim 3, characterized in that, When the current semantic state is a confirmed reservation state, it also includes: Within a preset grace period after the end of the appointment time window, the formula is used. Calculations are performed to obtain the updated initial dynamic threshold, where T is... S1 T represents the first preset value. S3 This represents the third preset value, t. end t represents the end time of the reservation time window. grace This indicates the preset grace period, where t represents the current time, and t end <t≤t end +t grace ; The updated initial dynamic threshold is limited to the preset numerical range to obtain the dynamic recognition threshold.

5. The intelligent management method for community public spaces according to claim 1, characterized in that, The step of determining whether the similarity score is greater than or equal to the dynamic recognition threshold includes: When the similarity score is less than the dynamic recognition threshold and the current semantic state is a confirmed reservation state, the alternative access process is triggered; wherein, triggering the alternative access process includes displaying a dynamic QR code bound to the current reservation on the doorplate terminal, scanning and verifying the dynamic QR code within a preset validity period, and outputting an unlocking command after successful verification.

6. The intelligent management method for community public spaces according to claim 1, characterized in that, After switching the target space from the pre-startup state to the usage state, the process further includes: When the target space is in use and no personnel are detected within a preset time period, the target space is switched from the use state to the temporary vacancy state, and an energy-saving control command is sent to the environmental equipment. The energy-saving control command includes adjusting the lighting brightness to a preset brightness and adjusting the air conditioning temperature to a preset temperature. When the reservation time window ends and personnel are detected leaving the site, the target space is switched from the usage state to the usage end state, and the environmental equipment is turned off or switched to deep energy-saving mode.

7. The intelligent management method for community public spaces according to claim 1, characterized in that, The step of generating a billing result based on the first timestamp and the second timestamp includes: The actual usage time is calculated using the first timestamp as the billing start time and the second timestamp as the billing end time. If the actual usage time does not exceed the reservation time window, the fee will be calculated at the normal unit price; otherwise, the overtime fee will be calculated at the overtime unit price, and the user's credit score will be updated based on the billing results. If a user does not generate a passage record within the scheduled time window and does not cancel the reservation in advance, it is determined as a no-show, and a penalty will be calculated and the corresponding points will be deducted from their credit score. The credit score is dynamically adjusted based on no-show behavior, lateness behavior, on-time performance behavior, and early cancellation behavior, and users are divided into different levels based on their credit score. Different user levels have different reservation priorities, cancellation grace periods, and fee deduction rights.

8. The intelligent management method for community public spaces according to claim 1, characterized in that, The method further includes: In response to an intercom call request triggered by a user on the doorplate terminal, the system queries the list of responsible administrators bound to the reservation records for the current time period. The list of responsible administrators includes at least one administrator account sorted by priority. Detect the online status of each administrator in the list of responsible administrators, and initiate a video call to any online administrator based on the online status; If all administrators in the current time period's list of responsible administrators are offline or unresponsive, a broadcast call will be initiated, and an emergency notification will be sent to the preset management personnel.

9. An intelligent management device for community public spaces, characterized in that, The device is used as a doorplate terminal in a community public space, and the device includes: The reservation receiving unit is used to receive and verify the signature reservation token issued by the cloud, and cache the reservation record carried by the signature reservation token. The signature reservation token is generated by the cloud after the user completes the reservation, and the reservation record includes the user identity, reservation time window and environmental device control parameters. A semantic parsing unit is used to determine the current semantic state based on the reservation time window, and to determine the reservation confidence score based on the current semantic state; An environmental pre-control unit is used to send a pre-start control command to the environmental equipment in the target space according to the preset lead time of the environmental equipment control parameters, so that the target space switches from an idle standby state to a ready-to-start state, wherein the preset lead time is dynamically configured according to the type of environmental equipment. The biometric processing unit is used to collect the user's biometric image in response to the user arrival event, process the biometric image, and obtain image quality score, liveness confidence score and similarity score; The dynamic decision unit is used to calculate a dynamic recognition threshold based on the current semantic state, the reservation confidence score, the image quality score, the liveness confidence score, and the acquired user's historical behavior risk score, and to determine whether the similarity score is greater than or equal to the dynamic recognition threshold. If so, it outputs an unlocking command and switches the target space from the pre-startup state to the usage state. The billing management unit is used to record the first timestamp of the user's first entry based on the unlocking command, and to record the second timestamp of the user's last exit when the user leaves the premises, and to generate a billing result based on the first timestamp and the second timestamp.

10. A doorplate terminal, characterized in that, The doorplate terminal includes a processor, a memory, and a computer program stored in the memory. When the computer program is executed by the processor, it implements the steps of the intelligent management method for community public space as described in any one of claims 1 to 8.

Images