Intelligent face recognition system based on Nan Mongolian access control system

Abstract

The invention discloses an intelligent face recognition system based on a swan-gap access control system. Comprising a multi-modal biological characteristic acquisition module, a self-adaptive illumination compensation module, an edge-cloud cooperative computing module, a behavior intention analysis module, a privacy compliance processing module, an anti-AI cheating protection module, a self-adaptive authority management module and an energy consumption optimization and self-maintenance module. And data communication and task scheduling among the modules are realized through a swan gap distributed soft bus, and the system has the beneficial effects that the security, efficiency and privacy protection capability of the system are remarkably improved through the multi-mode biological characteristic acquisition module, the dynamic illumination compensation module, the edge-cloud cooperative computing module and the like. The anti-AI cheating protection and the living body detection guarantee the authenticity of recognition, and the fake face attack is avoided; the energy consumption optimization and self-maintenance module prolongs the service life of hardware and reduces the long-term operation and maintenance cost; the self-adaptive authority management and emergency escape priority mechanism enhances the emergency response capability.

Description

Technical Field

[0001] The present invention relates to the field of face recognition technology, and specifically to a face intelligent recognition system based on the Hongmeng access control system. Background Art

[0002] With the rapid development of the intelligent society, access control systems have gradually upgraded from traditional card swiping and password verification to biometric technology. Among them, facial recognition has become the mainstream solution due to its contactless and convenient features. However, the existing facial recognition access control system still has many technical bottlenecks.

[0003] The recognition rate of existing facial recognition access control systems drops significantly under extreme lighting conditions (strong backlight, low illumination). Relying on a single visible light camera makes it difficult to stably obtain high-quality facial images. Although some solutions use infrared fill light or HDR technology, they lack hardware-algorithm coordinated optimization, resulting in insufficient robustness in dynamic scenes. Traditional liveness detection relies on motion commands (such as blinking and shaking heads), which can be easily bypassed by high-precision 3D-printed masks or Deepfake-generated faces. It lacks the ability to analyze behavioral intentions (such as stalking and abnormal micro-expressions) and cannot prevent social engineering attacks. In addition, most traditional systems directly store raw facial images or unencrypted features, which violates privacy regulations such as GDPR and poses a risk of data leakage. They lack dynamic desensitization and temporary identity mechanisms, and once biometric features are leaked, they cannot be revoked. To this end, we propose a facial intelligent recognition system based on the Hongmeng access control system. Summary of the Invention

[0004] The purpose of the present invention is to provide a face intelligent recognition system based on the Hongmeng access control system to solve the problems raised in the above background technology.

[0005] To achieve the above objectives, the intelligent facial recognition system based on the Hongmeng access control system includes a multimodal biometric feature acquisition module, an adaptive lighting compensation module, an edge-cloud collaborative computing module, a behavioral intention analysis module, a privacy compliance processing module, an anti-AI deception protection module, an adaptive permission management module, and an energy consumption optimization and self-maintenance module. Data communication and task scheduling between modules are achieved through the Hongmeng distributed soft bus. The multimodal biometric feature acquisition module is used to synchronously capture multi-dimensional biometric features of the human face through heterogeneous sensors; The adaptive illumination compensation module is used to dynamically adjust optical hardware parameters and optimize image quality in real time under extreme illumination conditions; The edge-cloud collaborative computing module is used to allocate computing tasks on demand, with the local end performing low-latency lightweight recognition and the cloud processing complex re-identification or big data comparison, balancing response speed and recognition accuracy; The behavioral intention analysis module is used to analyze the potential threat intention of a person approaching the access control through micro-expressions and gait trajectories; The privacy compliance processing module is used to embed a privacy protection mechanism in the feature extraction and storage stages; The anti-AI deception protection module is used to defend against fake face attacks generated by Deepfake, and dually ensures the authenticity of living bodies through frequency domain analysis and physiological signal verification; The adaptive authority management module is used to dynamically adjust the access authority level according to the real-time scenario; The energy consumption optimization and self-maintenance module is used to extend the hardware life and reduce the long-term operation and maintenance costs of the system through equipment health prediction and intelligent sleep mechanism.

[0006] Preferably, the multimodal biometric feature acquisition module includes a dynamic living body detection unit and a 3D structured light auxiliary unit; The dynamic living body detection unit is used to synchronously capture the image through the infrared camera and the visible light camera to determine whether it is a real human face; The 3D structured light auxiliary unit is used to obtain facial depth information through structured light projection.

[0007] Preferably, the adaptive illumination compensation module includes a polarized light enhancement unit and a low illumination HDR synthesis unit; The polarized light enhancement unit is used to control the angle of the polarization filter through the RetinexNet algorithm combined with the Hongmeng Hardware Abstraction Layer (HAL); The low-light HDR synthesis unit is used to realize multi-frame image fusion using the neural network HDRnet.

[0008] Preferably, the edge-cloud collaborative computing module includes a local lightweight inference unit and a cloud re-identification unit; The local lightweight inference unit is used to run a streamlined face detection model on the Hongmeng device; The cloud re-identification unit is used to upload the local uncertain results to the cloud for secondary verification.

[0009] Preferably, the behavior intention analysis module includes a micro-expression detection unit and a gait trajectory prediction unit; The micro-expression detection unit is used to analyze abnormal micro-expressions of people when approaching the access control based on the micro-expression recognition algorithm of optical flow; The gait trajectory prediction unit is used to predict whether the walking path is compliant through millimeter wave radar and LSTM model.

[0010] Preferably, the privacy compliance processing module includes a differential privacy desensitization unit and a temporary identity token unit; The differential privacy desensitization unit is used to add noise to the facial feature vector using the ε-differential privacy algorithm; The temporary identity token unit is used to generate a one-time pass credential through the HarmonyOS distributed identity authentication service.

[0011] Preferably, the anti-AI deception protection module includes a frequency domain feature analysis unit and a heartbeat association detection unit; The frequency domain feature analysis unit is used to detect frequency domain anomalies of AI-generated images through Fourier spectrum; The heartbeat association detection unit is used to generate a one-time pass certificate through the Hongmeng distributed identity authentication service.

[0012] Preferably, the adaptive authority management module includes a dynamic trust score evaluation unit and an emergency escape priority unit; The trust score evaluation unit is used to calculate the real-time credibility of personnel based on the Bayesian network; The emergency escape priority unit is used to respond to the Hongmeng emergency event service to force the opening of access control.

[0013] Preferably, the energy consumption optimization and self-maintenance module includes a device health prediction unit and an intermittent wake-up unit; The device health prediction unit is used to predict the hardware life through the LSTM model; The intermittent wake-up unit is used to achieve intelligent sleep using the Hongmeng low-power core and the PIR sensor.

[0014] According to any of the above methods, the face intelligent recognition system based on the Hongmeng access control system includes the following steps: Step 1: Acquire a facial image through a multimodal acquisition module, and generate a feature vector after illumination compensation and liveness detection; Step 2: Determine access rights by combining behavioral intent analysis with the anti-AI deception module; Step 3: Control access control based on privacy compliance processing results and dynamic trust scores; Step 4: Realize end-cloud collaborative computing and energy consumption optimization through HarmonyOS distributed task scheduling.

[0015] Compared with the prior art, the present invention has the following beneficial effects: Through the dual protection of multimodal biometric feature collection, dynamic liveness detection, and anti-AI deception protection module, the present invention can effectively prevent attacks with fake faces and ensure that only real people can pass through the access control. It not only improves recognition accuracy, but also prevents deception by AI technologies such as Deepfake.

[0016] The energy consumption optimization and self-maintenance module of the present invention can extend the hardware life and reduce long-term operation and maintenance costs through equipment health prediction and intelligent sleep mechanism. Especially in low-power mode, the system can work efficiently while reducing battery consumption.

[0017] The privacy compliance processing module of the present invention embeds technologies such as differential privacy and temporary identity tokens to ensure that the privacy protection of facial data complies with relevant regulatory requirements and effectively avoids personal privacy leakage during feature extraction and storage.

[0018] The adaptive permission management module of the present invention can dynamically adjust access permissions according to real-time situations, and combined with behavioral intention analysis and emergency escape priority mechanism, it can respond quickly in emergency situations, improve emergency handling efficiency and personnel safety.

[0019] Through the edge-cloud collaborative computing module, the present invention allows the local end to perform low-latency lightweight recognition tasks, while the cloud handles complex re-identification tasks, thereby optimizing response speed and recognition accuracy. This distributed computing method ensures the efficiency and real-time performance of the system while balancing the hardware burden. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Module flow chart of the present invention; Figure 2 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0021] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0022] See also Figure 1-Figure 2 The present invention provides a technical solution: a face intelligent recognition system based on the Hongmeng access control system, including a multimodal biometric feature acquisition module, an adaptive illumination compensation module, an edge-cloud collaborative computing module, a behavioral intention analysis module, a privacy compliance processing module, an anti-AI deception protection module, an adaptive permission management module, and an energy consumption optimization and self-maintenance module, and data communication and task scheduling between modules are realized through the Hongmeng distributed soft bus; The multimodal biometric acquisition module is used to synchronously capture multi-dimensional biometric features of the face through heterogeneous sensors; the adaptive illumination compensation module is used to dynamically adjust optical hardware parameters and optimize image quality in real time under extreme lighting conditions; the edge-cloud collaborative computing module is used to allocate computing tasks on demand, perform low-latency lightweight recognition locally, and process complex re-identification or big data comparison in the cloud, balancing response speed and recognition accuracy; the behavioral intention analysis module is used to analyze the potential threat intentions of people approaching the access control through micro-expressions and gait trajectories; the privacy compliance processing module is used to embed privacy protection mechanisms in the feature extraction and storage links; the anti-AI deception protection module is used to defend against fake face attacks generated by Deepfake, and dually ensure the authenticity of the live body through frequency domain analysis and physiological signal verification; the adaptive permission management module is used to dynamically adjust the access permission level according to the real-time scenario; the energy consumption optimization and self-maintenance module is used to extend the hardware life and reduce the long-term operation and maintenance costs of the system through equipment health prediction and intelligent sleep mechanism.

[0023] It should be noted that the system hardware of the present invention includes: Main control unit: Hongmeng development board (Hi3861 / Hi3516) and NPU accelerator chip (Ascend 310); Hongmeng development board is responsible for system scheduling, data fusion and communication management; NPU accelerator chip is used for local AI inference (such as face detection and liveness determination); Sensor group: dual camera module, 3D structured light module, millimeter wave radar, polarization filter (electrically adjustable) and PIR human body sensor; dual cameras are used for visible light and infrared liveness detection, with a MIPI-CSI interface; 3D structured light module is used to obtain depth information for anti-counterfeiting, with a USB 3.0 interface; millimeter wave radar is used to detect gait and heartbeat, with an SPI interface; polarization filter is used to dynamically adapt to strong light environments, with a GPIO+PWM interface; PIR human body sensor is used to trigger low-power wake-up, with an I2C interface; Communications and Storage: 5G / Wi-Fi 6, HarmonyOS distributed soft bus, and secure storage chip (SE); 5G / Wi-Fi 6 is used for cloud-based collaborative computing; the HarmonyOS distributed soft bus is used to enable data sharing between multiple devices; the secure storage chip (SE) is used to encrypt and store facial feature vectors; Software architecture and algorithm implementation: The Hongmeng operating system layer includes the kernel layer, service layer, and application layer. The kernel layer is the lightweight LiteOS-A kernel that supports real-time task scheduling. The service layer is used to call distributed data management, AI framework, and security engine. The application layer is the access control app and management backend. When in use, the facial image is first acquired through the multimodal acquisition module, and the feature vector is generated after illumination compensation and liveness detection. Then, the access permission is determined by combining behavioral intention analysis and anti-AI deception module. The access control switch is controlled based on the privacy compliance processing results and dynamic trust points. Finally, end-cloud collaborative computing and energy consumption optimization are achieved through HarmonyOS distributed task scheduling.

[0024] The multimodal biometric acquisition module includes a dynamic liveness detection unit and a 3D structured light auxiliary unit; the dynamic liveness detection unit is used to capture images synchronously through an infrared camera and a visible light camera to determine whether it is a real face; the 3D structured light auxiliary unit is used to obtain facial depth information through structured light projection.

[0025] It should be noted that the dynamic liveness detection unit uses infrared cameras and visible light cameras to capture images simultaneously. Based on the liveness detection algorithm of the Generative Adversarial Network (GAN), it distinguishes between printed photos or video attacks and determines whether they are real faces. The 3D structured light auxiliary unit uses the ToF (Time of Flight) depth perception algorithm, combined with the sensor driving framework of Hongmeng, to obtain facial depth information through structured light projection to prevent 2D forgery. The dynamic liveness detection unit uses two video streams: RGB and infrared. The algorithm uses a PatchGAN-based liveness discrimination model and outputs a liveness confidence score (0-1). The unit runs on the NPU after quantization, with a frame rate of 30 fps or higher. In the 3D structured light auxiliary unit, the input is: structured light point cloud data; the algorithm: ICP (Iterative Closest Point) matches the pre-stored 3D face model and rejects 2D plane attacks.

[0026] The adaptive illumination compensation module includes a polarized light enhancement unit and a low-light HDR synthesis unit; the polarized light enhancement unit is used to control the angle of the polarization filter through the RetinexNet algorithm combined with the HarmonyOS Hardware Abstraction Layer (HAL); the low-light HDR synthesis unit is used to use the neural network HDRnet to realize multi-frame image fusion.

[0027] It should be noted that the control process of the polarized light enhancement unit is: detect the ambient brightness through the light intensity sensor, and then adjust the polarizer angle through the Hongmeng HAL layer; finally, use RetinexNet to remove glare from the polarized image; the control process of the low-light HDR synthesis unit is: multi-frame RAW data input → HDRnet neural network synthesis → output balanced lighting image.

[0028] The edge-cloud collaborative computing module includes a local lightweight inference unit and a cloud-based re-identification unit; the local lightweight inference unit is used to run a streamlined face detection model on the HarmonyOS device; the cloud-based re-identification unit is used to upload local uncertain results to the cloud for secondary verification.

[0029] It should be noted that the local lightweight inference unit can quickly complete face detection, alignment and preliminary feature extraction on the Hongmeng device side. First, the lightweight MTCNN model is used to locate the face area (taking time ≤ 30ms), then MobileFaceNet outputs a 512-dimensional feature vector (taking time ≤ 50ms), and finally calculates the cosine similarity with the pre-stored feature library (1:1000 scale) of the end-side security chip (SE). If the similarity is ≥ the threshold and the liveness detection is passed, it is directly released; if the similarity is ∈ (0.7, 0.85) or the liveness result is questionable, cloud re-identification is triggered; if the similarity is <0.7, access is denied and the exception is recorded; the cloud re-identification unit can perform high-precision review of local uncertain candidate results, and the cloud runs a model trained with ResNet-100+ArcFace loss function, supporting comparison of tens of millions of face libraries; during data upload, the HarmonyOS distributed soft bus encrypts and transmits locally extracted 512-dimensional feature vectors, original image encrypted slices (optional, for manual review), device ID, timestamp and other metadata to the cloud. The cloud uses ResNet-100 to extract 1024-dimensional high-precision features for the second time, and then searches the cloud feature library (1:N, N≤10^6) for the top-5 similar results. Finally, the cloud video liveness detection model is called. If the highest similarity is ≥0.9, an authorization command is returned to the access control terminal; if the highest similarity is <0.9, a rejection command and a suspected identity alarm are returned.

[0030] The behavioral intention analysis module includes a micro-expression detection unit and a gait trajectory prediction unit; the micro-expression detection unit is used to analyze abnormal micro-expressions of people approaching the access control system based on the optical flow micro-expression recognition algorithm; the gait trajectory prediction unit is used to predict whether the walking path is compliant through millimeter-wave radar and LSTM model.

[0031] It should be noted that the micro-expression detection unit uses the following inputs: 5 consecutive frames of images from an RGB camera; the algorithm uses the optical flow method (Farneback) and lightweight CNN, and outputs a "nervous / normal" classification result; the gait trajectory prediction unit uses the following inputs: millimeter-wave radar point cloud sequence; the algorithm uses the LSTM trajectory prediction, and abnormal trajectories (such as wandering) trigger an alarm.

[0032] The privacy compliance processing module includes a differential privacy desensitization unit and a temporary identity token unit; the differential privacy desensitization unit is used to add noise to the facial feature vector using the ε-differential privacy algorithm; the temporary identity token unit is used to generate a one-time pass through the HarmonyOS distributed identity authentication service.

[0033] It should be noted that the differential privacy desensitization unit can add controllable noise to the facial feature vector to achieve "available but invisible" data protection. The temporary identity token unit can replace the original biometric features as a pass credential to achieve "de-identification" authorization. If user A passes through the office building by face swiping, the facial point cloud is first obtained through the 3D structured light camera, encrypted by the SE chip and transmitted to the NPU for processing. Then MobileFaceNet extracts the features → adds Laplace noise of ε=0.5 → generates the hash value "a1b2...", then the Hongmeng Hichain service issues a temporary token, which is received and verified by the access controller. Finally, the original feature memory is erased, and only the "noise version feature + hash" is retained for subsequent comparison (storage period ≤7 days). This module combines Hongmeng's native hardware-level security capabilities with the differential privacy algorithm to ensure recognition accuracy while achieving the "Privacy by Design" regulatory requirements.

[0034] The anti-AI deception protection module includes a frequency domain feature analysis unit and a heartbeat association detection unit; the frequency domain feature analysis unit is used to detect frequency domain anomalies of AI-generated images through Fourier spectrum; the heartbeat association detection unit is used to generate a one-time pass through the HarmonyOS distributed identity authentication service.

[0035] It should be noted that when using the frequency domain feature analysis unit, the FFT spectrum of the face ROI area is input, and then the frequency domain anomalies of the AI-generated image are detected through frequency domain bandpass filtering and One-Class SVM algorithm; in the heartbeat association detection unit, the millimeter wave radar micro-Doppler signal is first input, and then the PPG signal is extracted to verify whether the heart rate is within the normal human range.

[0036] The adaptive permission management module includes a dynamic trust score evaluation unit and an emergency escape priority unit; the trust score evaluation unit is used to calculate the real-time credibility of personnel based on the Bayesian network; the emergency escape priority unit is used to respond to the forced opening of access control by the Harmony emergency event service.

[0037] It should be noted that the dynamic trust score assessment unit calculates personnel credibility based on multi-dimensional behavioral data and dynamically adjusts the access permission level; the emergency escape priority unit can forcibly lift access control restrictions and initiate escape guidance in emergency situations such as fire and earthquake; the workflow is as follows: when a high-risk person is dynamically intercepted, if person B approaches the confidential area during non-working hours, the facial matching degree is 0.88 (normal), but the micro-expression analysis shows "nervous" (probability 72%), the Bayesian network comprehensive calculation trust score = 58 points (<60 points threshold), the system automatically triggers the access control to remain closed, and pushes an alarm information (including real-time video stream) to the security personnel, and records the event in the audit log; when a fire emergency response is carried out, the fire sensor triggers the SIGABRT signal → the Harmony event bus broadcasts to all access control terminals, and the emergency unit completes the release of the electromagnetic lock, shuts down the facial recognition process to save computing power, and starts emergency lighting and voice guidance within 100ms, and the escape personnel count result is displayed in real time on the large screen in the fire control room.

[0038] The energy consumption optimization and self-maintenance module includes a device health prediction unit and an intermittent wake-up unit; the device health prediction unit is used to predict the hardware life through the LSTM model; the intermittent wake-up unit is used to use the HarmonyOS low-power core and PIR sensor to achieve intelligent sleep.

[0039] It should be noted that the device health prediction unit can predict the remaining life of key hardware components based on the LSTM timing model, and provide early warning of failures; the intermittent wake-up unit can reduce the device's standby power consumption to less than 20% of traditional solutions through an intelligent sleep strategy; during daily low-power operation, the system enters deep sleep 5 minutes after the person leaves, and only the PIR sensor and coprocessor remain powered. When a person approaches within 3 meters, the PIR triggers an interrupt → the Hongmeng kernel completes the startup of the main control chip, activation of the RGB camera, and loading of the face detection model to the NPU within 100ms. If the recognition timeout (no face within 30 seconds), it automatically returns to low-power monitoring mode.

[0040] In summary: The present invention significantly improves the security, efficiency and privacy protection capabilities of the system through modules such as multimodal biometric feature acquisition, dynamic lighting compensation, and edge-cloud collaborative computing. Its anti-AI deception protection and liveness detection ensure the authenticity of recognition and avoid fake face attacks; energy consumption optimization and self-maintenance modules extend the service life of hardware and reduce long-term operation and maintenance costs; adaptive permission management and emergency escape priority mechanism enhance emergency response capabilities; and privacy compliance processing ensures data compliance and security. In addition, the system's flexibility and efficient collaborative computing capabilities further improve recognition accuracy and response speed, making access control management more intelligent and reliable.

[0041] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. The intelligent face recognition system based on Hongmeng access control system is characterized by: It includes a multimodal biometric acquisition module, an adaptive illumination compensation module, an edge-cloud collaborative computing module, a behavioral intention analysis module, a privacy compliance processing module, an anti-AI deception protection module, an adaptive permission management module, and an energy consumption optimization and self-maintenance module. Data communication and task scheduling between modules are achieved through the Hongmeng distributed soft bus. The multimodal biometric feature acquisition module is used to synchronously capture multi-dimensional biometric features of the human face through heterogeneous sensors; The adaptive illumination compensation module is used to dynamically adjust optical hardware parameters and optimize image quality in real time under extreme illumination conditions; The edge-cloud collaborative computing module is used to allocate computing tasks on demand, with the local end performing low-latency lightweight recognition and the cloud processing complex re-identification or big data comparison, balancing response speed and recognition accuracy; The behavioral intention analysis module is used to analyze the potential threat intention of a person approaching the access control through micro-expressions and gait trajectories; The privacy compliance processing module is used to embed a privacy protection mechanism in the feature extraction and storage stages; The anti-AI deception protection module is used to defend against fake face attacks generated by Deepfake, and dually ensures the authenticity of living bodies through frequency domain analysis and physiological signal verification; The adaptive authority management module is used to dynamically adjust the access authority level according to the real-time scenario; The energy consumption optimization and self-maintenance module is used to extend the hardware life and reduce the long-term operation and maintenance costs of the system through equipment health prediction and intelligent sleep mechanism.

2. The face intelligent recognition system based on the Hongmeng access control system according to claim 1 is characterized in that: The multimodal biometric feature acquisition module includes a dynamic living body detection unit and a 3D structured light auxiliary unit; The dynamic living body detection unit is used to synchronously capture the image through the infrared camera and the visible light camera to determine whether it is a real human face; The 3D structured light auxiliary unit is used to obtain facial depth information through structured light projection.

3. The face intelligent recognition system based on the Hongmeng access control system according to claim 1 is characterized in that: The adaptive illumination compensation module includes a polarized light enhancement unit and a low illumination HDR synthesis unit; The polarized light enhancement unit is used to control the angle of the polarization filter through the RetinexNet algorithm combined with the Hongmeng Hardware Abstraction Layer (HAL); The low-light HDR synthesis unit is used to realize multi-frame image fusion using the neural network HDRnet.

4. The face intelligent recognition system based on the Hongmeng access control system according to claim 1 is characterized in that: The edge-cloud collaborative computing module includes a local lightweight inference unit and a cloud re-identification unit; The local lightweight inference unit is used to run a streamlined face detection model on the Hongmeng device; The cloud re-identification unit is used to upload the local uncertain results to the cloud for secondary verification.

5. The face intelligent recognition system based on the Hongmeng access control system according to claim 1 is characterized in that: The behavior intention analysis module includes a micro-expression detection unit and a gait trajectory prediction unit; The micro-expression detection unit is used to analyze abnormal micro-expressions of people when approaching the access control based on the micro-expression recognition algorithm of optical flow; The gait trajectory prediction unit is used to predict whether the walking path is compliant through millimeter wave radar and LSTM model.

6. The face intelligent recognition system based on the Hongmeng access control system according to claim 1 is characterized in that: The privacy compliance processing module includes a differential privacy desensitization unit and a temporary identity token unit; The differential privacy desensitization unit is used to add noise to the facial feature vector using the ε-differential privacy algorithm; The temporary identity token unit is used to generate a one-time pass credential through the HarmonyOS distributed identity authentication service.

7. The face intelligent recognition system based on the Hongmeng access control system according to claim 1 is characterized in that: The anti-AI deception protection module includes a frequency domain feature analysis unit and a heartbeat association detection unit; The frequency domain feature analysis unit is used to detect frequency domain anomalies of AI-generated images through Fourier spectrum; The heartbeat association detection unit is used to generate a one-time pass certificate through the Hongmeng distributed identity authentication service.

8. The face intelligent recognition system based on the Hongmeng access control system according to claim 1 is characterized in that: The adaptive rights management module includes a dynamic trust score evaluation unit and an emergency escape priority unit; The trust score evaluation unit is used to calculate the real-time credibility of personnel based on the Bayesian network; The emergency escape priority unit is used to respond to the Hongmeng emergency event service to force the opening of access control.

9. The face intelligent recognition system based on the Hongmeng access control system according to claim 1 is characterized in that: The energy consumption optimization and self-maintenance module includes a device health prediction unit and an intermittent wake-up unit; The device health prediction unit is used to predict the hardware life through the LSTM model; The intermittent wake-up unit is used to achieve intelligent sleep using the Hongmeng low-power core and the PIR sensor.

10. The face intelligent recognition system method based on the Hongmeng access control system according to any one of claims 1 to 9 is characterized in that: The steps include: Step 1: Acquire a facial image through a multimodal acquisition module, and generate a feature vector after illumination compensation and liveness detection; Step 2: Determine access rights by combining behavioral intent analysis with the anti-AI deception module; Step 3: Control access control based on privacy compliance processing results and dynamic trust scores; Step 4: Realize end-cloud collaborative computing and energy consumption optimization through HarmonyOS distributed task scheduling.

Images