Intelligent door lock real-time visual analysis method and system based on edge calculation

Abstract

The invention discloses an intelligent door lock real-time visual analysis method and system based on edge calculation, and relates to the technical field of door locks, and the method comprises the steps: collecting a user face image in real time through a multi-mode sensor integrated with an intelligent door lock, and synchronously obtaining a distance value and dynamic motion information between a face and a camera; judging whether a living body detection and recognition process is triggered or not; when a living body detection and recognition process is triggered, extracting image features, depth information and Doppler frequency shift dynamic parameters of a face region; generating a composite living body feature vector containing a three-dimensional structure, texture details and motion characteristics; and performing dual verification on the generated composite living body feature vector. The method has the advantages that by introducing the Doppler frequency shift detection technology of the millimeter wave radar, the specific dynamic micro-motion characteristics of the living body are creatively fused into the living body authentication system, the defect that traditional living body detection excessively depends on static textures is effectively overcome, and the biological motion fingerprint authentication capacity of the intelligent door lock is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of door locks, and in particular to a real-time visual analysis method and system for smart door locks based on edge computing. Background Art

[0002] With the rapid development of smart security technology, biometric-based smart door locks have become widely used in the smart home sector. Traditional solutions often rely on a single visual sensor for two-dimensional facial recognition, making them vulnerable to spoofing attacks such as high-precision photos and video replays. Furthermore, static feature extraction struggles to protect against new attack methods, such as 3D-printed masks. While existing technologies employ infrared cameras or structured light for liveness detection, they carry the risk of misjudging depth information in complex lighting environments. Furthermore, liveness verification methods that rely on cloud computing suffer from high network latency and privacy data leakage, making them unable to meet the dual real-time and security requirements of door locks.

[0003] Current smart door lock systems face significant technical bottlenecks in the fusion of dynamic liveness features: on the one hand, vision-based micro-expression detection is susceptible to interference from environmental noise, making it difficult to distinguish between real biological movement and mechanical bionic movements; on the other hand, the lack of collaborative perception of multi-dimensional spatial information between the user and the device allows attackers to evade detection by adjusting the distance of the forged object. In addition, existing edge computing architectures often use simple cascade processing for feature fusion, failing to effectively explore the correlation of dynamic features in the spatiotemporal domain, resulting in insufficient confidence in liveness judgment. These shortcomings severely restrict the application reliability of smart door locks in high-security scenarios, and a new solution that can integrate multimodal biometric features and possess localized real-time analysis capabilities is urgently needed. Summary of the Invention

[0004] In order to solve the above technical problems, a real-time visual analysis method and system for smart door locks based on edge computing are provided. This technical solution solves the problem in the existing technology that vision-based micro-expression detection is easily interfered by environmental noise and has difficulty in distinguishing between real biological movements and mechanical bionic movements.

[0005] In order to achieve the above purpose, the technical solution adopted by the present invention is: A real-time visual analysis method for smart door locks based on edge computing, including: The multimodal sensor integrated in the smart door lock collects the user's facial image in real time, and simultaneously obtains the distance between the face and the camera and dynamic motion information, which includes Doppler frequency shift characteristics; Based on the distance value and the Doppler frequency shift characteristics fed back by the millimeter-wave radar, the edge computing node analyzes the frequency of the user's facial micro-movements to determine whether to trigger the liveness detection and recognition process; When the liveness detection and recognition process is triggered, the edge computing node is used to perform local processing on the collected facial images to extract the image features, depth information, and Doppler frequency motion parameters of the facial area; The edge-side lightweight model is used to perform spatiotemporal fusion of image features, depth information, and Doppler shift features to generate a composite living feature vector containing three-dimensional structure, texture details, and motion characteristics. The generated composite living body feature vector is matched and retrieved with a registered user feature library pre-stored in an edge storage device. The feature library contains encrypted user three-dimensional topology structure templates and Doppler frequency shift dynamic baseline data. During the matching process, dual verification is performed by combining static feature similarity and dynamic frequency shift trajectory consistency; If the verification is successful, the door opening action is triggered. If the verification fails, a local alarm is triggered through the edge node and access is denied. At the same time, the abnormal characteristics are encrypted and uploaded to the cloud for threat analysis.

[0006] Preferably, the process of determining whether to trigger liveness detection and identification specifically includes: The millimeter-wave radar transmits a frequency-modulated continuous wave and calculates the precise distance value based on the phase difference of the echo signal; The millimeter-wave radar is used to capture the frequency shift signal caused by micro-movements on the surface of the human face in real time, and the frequency shift amplitude and phase change data are extracted; Perform time-frequency analysis on the frequency-shifted signal based on edge computing nodes to calculate the periodic motion characteristics generated by natural breathing or micro-expressions on the face; If the detected distance value is within the set range but the Doppler shift feature is missing, it is determined to be a static forgery attack and an alarm is triggered directly; If the detected periodic motion characteristics match the preset living body biometric characteristics and the distance value to the door lock is less than the detection threshold, the living body detection and recognition process is activated.

[0007] Furthermore, a real-time visual analysis system for smart door locks based on edge computing is proposed, including: A multimodal data acquisition module, integrated into the smart door lock, includes a camera and millimeter-wave radar. It is used to capture the user's facial image in real time, simultaneously obtain the distance between the face and the camera, and extract the Doppler shift characteristics of dynamic motion information based on the millimeter-wave radar echo signal. The edge computing trigger module is deployed on the edge computing node and includes a distance analysis unit, a frequency shift analysis unit, and a liveness determination unit. The distance analysis unit calculates the precise distance value based on the millimeter-wave radar phase difference. The frequency shift analysis unit extracts the periodic motion characteristics of facial micro-movements through time-frequency analysis. The liveness determination unit activates the liveness detection process when it detects that the distance value is within a set range and the multi-shift characteristics match the liveness biometric characteristics. A local feature processing module includes an image feature extraction unit, a depth information registration unit, and a dynamic trajectory encoding unit. The image feature extraction unit generates a two-dimensional texture feature map through a convolutional neural network, the depth information registration unit constructs a three-dimensional topological structure of the face, and the dynamic trajectory encoding unit converts Doppler frequency shift signals into time series features. The spatiotemporal fusion module uses a lightweight model and attention mechanism to perform cross-modal weighted fusion of 2D texture feature maps, 3D topological structures, and time series features to generate a composite living feature vector that includes 3D structure, texture details, and motion characteristics. A dual verification module, connected to the encrypted user feature library in the edge storage device, includes a static feature matching unit and a dynamic trajectory verification unit. The static feature matching unit calculates the similarity between the three-dimensional topology structure and the registration template, and the dynamic trajectory verification unit aligns the real-time frequency shift trajectory with the baseline data through dynamic time warping and verifies the consistency of the spectrum biological regularity. The security execution module includes a door lock control unit, a local alarm unit and an encryption upload unit. The door lock control unit triggers the door opening action when the double verification is passed. The local alarm unit triggers an alarm when the verification fails or a static forgery attack is detected. The encryption upload unit desensitizes and homomorphically encrypts the abnormal features and then uploads them to the cloud threat analysis platform.

[0008] Compared with the prior art, the present invention has the following beneficial effects: By introducing the Doppler shift detection technology of millimeter-wave radar, this invention pioneered the integration of dynamic micro-motion characteristics unique to living organisms into a liveness authentication system. Doppler shift features can accurately capture millimeter-level micro-motion signals on a user's face caused by breathing, micro-expressions, and blood circulation. These biometric features are difficult to simulate using 3D-printed masks, dynamic video playback, or mechanical drive devices, effectively addressing the traditional liveness detection's over-reliance on static textures. By analyzing the phase changes and periodicity of the frequency-shifted signal in real time, the system can distinguish between authentic skin tissue vibrations and the mechanical vibration patterns of counterfeit attacks. Combined with the spatiotemporal fusion of three-dimensional structural features, the system significantly improves its defense accuracy against novel attack methods such as air flow artifacts inside silicone masks and high-frame-rate screen flicker. Furthermore, the dynamic baseline verification mechanism of Doppler shift significantly reduces the impact of ambient light interference on liveness determination, providing contactless door locks with reliable, all-weather biometric motion fingerprint authentication capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] Figure 1 This is a flow chart of the real-time visual analysis method for smart door locks based on edge computing proposed in Example 1; Figure 2 This is a flow chart of the method for determining whether to trigger the liveness detection and identification process proposed in Example 1; Figure 3 This is a flow chart of the method for generating a composite living body feature vector including three-dimensional structure, texture details and motion characteristics proposed in Example 2; Figure 4 Schematic diagram of the computer-readable storage medium structure proposed for this solution. DETAILED DESCRIPTION

[0010] The following description is intended to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are merely examples, and those skilled in the art may conceive of other obvious variations.

[0011] Example 1:

[0012] Reference Figure 1 As shown in FIG, a real-time visual analysis method for smart door locks based on edge computing includes: The multimodal sensor integrated in the smart door lock collects the user's facial image in real time, and simultaneously obtains the distance between the face and the camera and dynamic motion information, which includes Doppler frequency shift characteristics; The multimodal sensor is composed of a high-resolution camera and a millimeter-wave radar. The millimeter-wave radar transmits a 60GHz frequency-modulated continuous wave (FMCW) signal, using the phase difference of the echo signal to calculate submillimeter distance accuracy. It also uses the Doppler effect to capture facial skin micro-vibration signals in the 0.1-5Hz range. This design overcomes the technical limitations of traditional visual sensors that separate distance perception and motion monitoring, achieving simultaneous calibration of spatial position and biological dynamic characteristics. Based on the distance value and the Doppler frequency shift characteristics fed back by the millimeter-wave radar, the edge computing node analyzes the frequency of the user's facial micro-movements to determine whether to trigger the liveness detection and recognition process; The edge computing node has a built-in adaptive threshold decision algorithm that dynamically adjusts the trigger conditions for liveness detection by jointly analyzing the distance attenuation curve (such as the user's approach speed) and the Doppler spectrum energy distribution (such as the fundamental breathing frequency of 0.8-2.4Hz). When non-biological frequency shift characteristics (such as the high-frequency component of the mechanical vibration of a silicone mask) or abnormal approach behavior (such as sudden acceleration) are detected, the defense protocol is immediately activated to effectively block brute force attacks. When the liveness detection and recognition process is triggered, the edge computing node is used to perform local processing on the collected facial images to extract the image features, depth information, and Doppler frequency motion parameters of the facial area; The localization processing adopts a staged feature distillation technology. First, the lightweight MobileNet-V3 network is used to extract the heat map of facial key points. Then, the millimeter-wave radar point cloud data is used to generate a three-dimensional curvature map with millimeter-level accuracy. At the same time, the 0.5-3Hz micro-Doppler modulation component caused by capillary pulsation is separated from the original frequency shift signal.

[0013] The edge-side lightweight model is used to perform spatiotemporal fusion of image features, depth information, and Doppler shift features to generate a composite living feature vector containing three-dimensional structure, texture details, and motion characteristics. The spatiotemporal domain fusion adopts a multi-branch Transformer architecture, in which the spatial branch performs graph convolution encoding on the three-dimensional point cloud, the temporal branch performs LSTM modeling on the Doppler frequency shift sequence, and realizes the correlation analysis between skin texture vibration and facial muscle movement through the cross-modal attention mechanism.

[0014] The generated composite living body feature vector is matched and retrieved with a registered user feature library pre-stored in an edge storage device. The feature library contains encrypted user three-dimensional topology structure templates and Doppler frequency shift dynamic baseline data. During the matching process, dual verification is performed by combining static feature similarity and dynamic frequency shift trajectory consistency; The registered user feature database is constructed in the following way: During the user registration phase, millimeter-wave radar is used to collect multiple sets of dynamic frequency shift baseline data, including natural head movements, blinking, and breathing-related signals; After denoising the original frequency shift data, the frequency shift amplitude mean and phase change variance are extracted as dynamic baseline features; The dynamic baseline features are bound to the three-dimensional face template and stored in the trusted execution environment of the edge storage device after encryption.

[0015] If the verification is successful, the door opening action is triggered. If the verification fails, a local alarm is triggered through the edge node and access is denied. At the same time, the abnormal characteristics are encrypted and uploaded to the cloud for threat analysis.

[0016] Static matching is based on a modified cosine similarity metric that incorporates depth perception, while dynamic verification uses a dynamic time warping (DTW) algorithm to calculate the morphological similarity of frequency-shifted trajectories. Differential privacy is implemented during the matching process to ensure that the biometric template is irreversible and cannot be reversed.

[0017] Reference Figure 2 As shown, in this embodiment, it is further proposed that the process of determining whether to trigger liveness detection and identification specifically includes: The millimeter wave radar transmits frequency modulated continuous waves and calculates the precise distance value based on the time difference of the echo signal; Based on the time when the millimeter wave signal is sent and received, and the signal propagation speed, the distance d of the noise source is calculated. , where C is the signal propagation speed, is the speed of light, and t is the time between the millimeter wave signal being sent and received; The millimeter-wave radar is used to capture the frequency shift signal caused by micro-movements on the surface of the human face in real time, and the frequency shift amplitude and phase change data are extracted; Perform time-frequency analysis on the frequency-shifted signal based on edge computing nodes to calculate the periodic motion characteristics generated by natural breathing or micro-expressions on the face; The calculation formula based on Doppler frequency shift is: ,

[0018] in, is the frequency of the echo signal received by the millimeter-wave radar, is the signal frequency emitted by the millimeter wave radar, is the signal propagation speed, is the speed of light, The relative motion speed between the face surface and the millimeter-wave radar caused by micro-movement of the face surface; Based on this, by analyzing the frequency shift signal between the echo signal frequency received by the millimeter-wave radar and the signal frequency emitted by the millimeter-wave radar, the living biological characteristics caused by the micro-movement of the human face surface can be obtained; By analyzing the time-frequency characteristics of the frequency-shifted signal, the periodic motion characteristics of the human body's natural breathing or micro-expressions can be obtained; If the detected distance value is within the set range but the Doppler shift feature is missing, it is determined to be a static forgery attack and an alarm is triggered directly; If the Doppler shift feature is not detected, but a face to be identified is actually detected in the area, it can be determined that a non-live face exists. In this case, it can be identified as a static forgery attack and an alarm will be triggered directly; If the detected periodic motion characteristics match the preset living body biometric characteristics and the distance value to the door lock is less than the detection threshold, the living body detection and recognition process is activated.

[0019] If and only if there is a face in the door lock determination area and it is alive, the subsequent facial feature recognition and matching process will be carried out.

[0020] Example 2:

[0021] Reference Figure 3 As shown, based on the first embodiment, this embodiment proposes: using the edge-side lightweight model to perform spatiotemporal fusion of image features, depth information, and Doppler shift features to generate a composite living feature vector containing three-dimensional structure, texture details, and motion characteristics. The specific steps are as follows: Perform convolutional neural network extraction on image features to generate a two-dimensional texture feature map; Two-dimensional processing uses convolutional neural networks to extract feature vectors, including geometric features of facial key points such as eyebrow distance and nose bridge curvature; Perform point cloud registration on depth information to construct the 3D topological structure of the face; 3D processing uses point cloud registration. Based on the original depth map obtained by the ToF camera, the statistical outlier removal (SOR) algorithm is used to filter out noise points. Combined with Poisson surface reconstruction, a continuous 3D mesh model is generated. 200 topological feature points such as the nose tip and eye sockets are extracted, retaining three-dimensional features such as nose tip height and eye socket depth. The 3D topological structure of the face with normal and curvature information is constructed. The Doppler frequency shift trajectory is encoded as a time series feature and then weightedly fused with the two-dimensional texture feature map and the three-dimensional topological structure using an attention mechanism.

[0022] The Doppler frequency shift signal of the millimeter-wave radar is converted into a time-frequency spectrum through STFT, and the time-domain motion features are extracted through 1D convolution; Based on the cross-modal multi-head attention mechanism, the time domain features are used as query vectors, and the texture and 3D topological features are used as key-value pairs to calculate the space-time association weights. Finally, by performing temporal fusion on the weighted features, a composite living feature vector is output to synchronously characterize the texture, structure and biological micro-motion characteristics.

[0023] Specifically, the dual verification of the combined static feature similarity and dynamic frequency shift trajectory consistency includes: The similarity between the three-dimensional topological structure and the template exceeds a first threshold; The similarity between the 3D topological structure and the template is measured using cosine similarity, and deep matching is introduced. The specific steps include: The depth data of the topological feature points in the three-dimensional topological structure are compared with the depth matrix stored in the database to calculate the mean square error, and the templates with a mean square error less than 0.1 are selected as candidate templates; Calculate the cosine similarity between the texture feature vector in the three-dimensional topological structure and the reference vector of the candidate template stored in the database; If the cosine similarity is greater than a first threshold value, which is usually set to 0.85, then it is determined that the similarity verification between the three-dimensional topological structure and the template has passed; The deviation of the Doppler frequency motion trajectory from the baseline is lower than a second threshold; The deviation of the Doppler frequency dynamic trajectory from the baseline is calculated using DTW, which includes: Perform dynamic time warping alignment on the real-time frequency shift trajectory and the baseline data, and calculate the minimum path distance between the trajectories; If the path distance exceeds the second threshold, it is determined to be a fake motion mode; Z-score normalization is used to normalize the Doppler frequency moving trajectory and the baseline stored in the database; Calculate the Euclidean distance between the Doppler frequency shift trajectory and all the point pairs in the baseline stored in the database; Find a path with the minimum cumulative distance. The cumulative distance of the path is the DTW distance between the Doppler frequency moving trajectory and the baseline stored in the database. The DTW distance is used as the deviation between the Doppler frequency moving trajectory and the baseline. The smaller the DTW distance, the more similar the curve shape is.

[0024] If both of the above conditions are met, the verification is successful.

[0025] Example 3:

[0026] Based on the second embodiment, this embodiment further introduces the liveness detection confidence level, which is used to analyze whether the acquired Doppler frequency shift trajectory is caused by the movement of the living body. Specifically, it includes: The continuous Doppler frequency shift trajectory is sliced ​​according to the time window with an overlap rate of 50% to ensure that the complete cycle of the periodic motion is captured and several signal segments are obtained; Perform STFT on each signal segment to generate a time-frequency spectrum. Identify the frequency component with the largest amplitude in the spectrum to verify whether it falls within the frequency band characteristic of living organisms. If so, detect whether it is accompanied by a second harmonic. For example, if the frequency component with the largest amplitude in the spectrum is 0.15Hz, it falls within the frequency band characteristic of breathing. In this case, detect whether there is an energy peak at 0.3Hz. The autocorrelation function is used to verify whether the signal is periodic, and the ACF peak interval of body respiration / heartbeat is stable; The facial micro-movements detected by the camera are synchronously compared with the energy distribution of the corresponding frequency band of the Doppler spectrum, such as the blinking frequency of 0.1-0.3Hz. If a breathing signal is detected but there is no facial micro-movement, the person is judged as non-living.

[0027] Example 4:

[0028] Based on the same inventive concept as the above-mentioned embodiments 1 to 3, this solution also proposes a real-time visual analysis system for smart door locks based on edge computing, including: A multimodal data acquisition module, integrated into the smart door lock, includes a camera and millimeter-wave radar. It is used to capture the user's facial image in real time, simultaneously obtain the distance between the face and the camera, and extract the Doppler shift characteristics of dynamic motion information based on the millimeter-wave radar echo signal. The edge computing trigger module is deployed on the edge computing node and includes a distance analysis unit, a frequency shift analysis unit, and a liveness determination unit. The distance analysis unit calculates the precise distance value based on the millimeter-wave radar phase difference. The frequency shift analysis unit extracts the periodic motion characteristics of facial micro-movements through time-frequency analysis. The liveness determination unit activates the liveness detection process when it detects that the distance value is within a set range and the multi-shift characteristics match the liveness biometric characteristics. A local feature processing module includes an image feature extraction unit, a depth information registration unit, and a dynamic trajectory encoding unit. The image feature extraction unit generates a two-dimensional texture feature map through a convolutional neural network, the depth information registration unit constructs a three-dimensional topological structure of the face, and the dynamic trajectory encoding unit converts Doppler frequency shift signals into time series features. The spatiotemporal fusion module uses a lightweight model and attention mechanism to perform cross-modal weighted fusion of 2D texture feature maps, 3D topological structures, and time series features to generate a composite living feature vector that includes 3D structure, texture details, and motion characteristics. A dual verification module, connected to the encrypted user feature library in the edge storage device, includes a static feature matching unit and a dynamic trajectory verification unit. The static feature matching unit calculates the similarity between the three-dimensional topology structure and the registration template, and the dynamic trajectory verification unit aligns the real-time frequency shift trajectory with the baseline data through dynamic time warping and verifies the consistency of the spectrum biological regularity. The security execution module includes a door lock control unit, a local alarm unit and an encryption upload unit. The door lock control unit triggers the door opening action when the double verification is passed. The local alarm unit triggers an alarm when the verification fails or a static forgery attack is detected. The encryption upload unit desensitizes and homomorphically encrypts the abnormal features and then uploads them to the cloud threat analysis platform.

[0029] Figure 4 This is a schematic diagram of the computer-readable storage medium structure provided by an embodiment of the present application. Figure 4 1 shows a computer-readable storage medium 600 according to one embodiment of the present application. Computer-readable instructions are stored on the computer-readable storage medium 600. When the computer-readable instructions are executed by a processor, the quantitative analysis method for evaluating hot dry rock targets according to the embodiment of the present application described with reference to the above figures can be executed. The storage medium 600 includes, but is not limited to, volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and cache memory. Non-volatile memory may include, for example, read-only memory (ROM), a hard disk, flash memory, etc.

[0030] In summary, the advantages of the present invention lie in its pioneering integration of the unique dynamic micro-motion characteristics of living organisms into a liveness authentication system by introducing millimeter-wave radar's Doppler shift detection technology. Doppler shift features can accurately capture millimeter-level micro-motion signals on a user's face caused by breathing, micro-expressions, and blood circulation. These biometric features are difficult to simulate using 3D-printed masks, dynamic video playback, or mechanical drive devices, effectively addressing the drawback of traditional liveness detection's over-reliance on static textures. By analyzing the phase changes and periodicity of the frequency-shifted signal in real time, the system can distinguish between genuine skin tissue vibrations and the mechanical vibration patterns of counterfeit attacks. Combined with the spatiotemporal fusion of three-dimensional structural features, the system significantly improves its defense accuracy against novel attack methods such as air flow artifacts inside silicone masks and high-frame-rate screen flicker. Furthermore, the dynamic baseline verification mechanism of Doppler shift significantly reduces the impact of ambient light interference on liveness determination, providing contactless door locks with reliable, all-weather biometric motion fingerprint authentication capabilities.

[0031] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions merely illustrate the principles of the present invention. Various changes and modifications may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and modifications are intended to fall within the scope of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

[0032] It should be noted that references in the specification to "one embodiment," "an embodiment," "exemplary embodiments," "some embodiments," etc. indicate that the described embodiments may include specific features, structures, or characteristics, but not necessarily every embodiment will include such specific features, structures, or characteristics. Furthermore, when specific features, structures, or characteristics are described in conjunction with an embodiment, it is within the knowledge of persons skilled in the relevant art to implement such features, structures, or characteristics in conjunction with other embodiments (whether or not explicitly described).

[0033] In general, terms can be understood, at least in part, from their use in context. For example, depending at least in part on the context, the term "one or more" as used herein can be used to describe any feature, structure, or characteristic in the singular sense, or can be used to describe a combination of features, structures, or characteristics in the plural sense. Additionally, the term "based on" can be understood as not necessarily intended to convey an exclusive set of factors, but can instead, depending at least in part on the context, allow for the presence of other factors that are not necessarily explicitly described.

Claims

1. A real-time visual analysis method for smart door locks based on edge computing, characterized in that: include: The multimodal sensor integrated in the smart door lock collects the user's facial image in real time, and simultaneously obtains the distance between the face and the camera and dynamic motion information, which includes Doppler frequency shift characteristics; Based on the distance value and the Doppler frequency shift characteristics fed back by the millimeter-wave radar, the edge computing node analyzes the frequency of the user's facial micro-movements to determine whether to trigger the liveness detection and recognition process; When the liveness detection and recognition process is triggered, the edge computing node is used to perform local processing on the collected facial images to extract the image features, depth information, and Doppler frequency motion parameters of the facial area; The edge-side lightweight model is used to perform spatiotemporal fusion of image features, depth information, and Doppler shift features to generate a composite living feature vector containing three-dimensional structure, texture details, and motion characteristics. The generated composite living body feature vector is matched and retrieved with a registered user feature library pre-stored in an edge storage device. The feature library contains encrypted user three-dimensional topology structure templates and Doppler frequency shift dynamic baseline data. During the matching process, dual verification is performed by combining static feature similarity and dynamic frequency shift trajectory consistency; If the verification is successful, the door opening action is triggered. If the verification fails, a local alarm is triggered through the edge node and access is denied. At the same time, the abnormal characteristics are encrypted and uploaded to the cloud for threat analysis.

2. A real-time visual analysis method for smart door locks based on edge computing according to claim 1, characterized in that: The process of determining whether to trigger liveness detection and identification specifically includes: The millimeter-wave radar transmits a frequency-modulated continuous wave and calculates the precise distance value based on the phase difference of the echo signal; The millimeter-wave radar is used to capture the frequency shift signal caused by micro-movements on the surface of the human face in real time, and the frequency shift amplitude and phase change data are extracted; Perform time-frequency analysis on the frequency-shifted signal based on edge computing nodes to calculate the periodic motion characteristics generated by natural breathing or micro-expressions on the face; If the detected distance value is within the set range but the Doppler shift feature is missing, it is determined to be a static forgery attack and an alarm is triggered directly; If the detected periodic motion characteristics match the preset living body biometric characteristics and the distance value to the door lock is less than the detection threshold, the living body detection and recognition process is activated.

3. The real-time visual analysis method for smart door locks based on edge computing according to claim 2 is characterized in that: The method of performing spatiotemporal fusion of image features, depth information, and Doppler shift features through the edge-side lightweight model to generate a composite living feature vector containing three-dimensional structure, texture details, and motion characteristics specifically includes: Perform convolutional neural network extraction on image features to generate a two-dimensional texture feature map; Perform point cloud registration on depth information to construct the 3D topological structure of the face; The Doppler frequency shift trajectory is encoded as a time series feature and then weightedly fused with the two-dimensional texture feature map and the three-dimensional topological structure using an attention mechanism.

4. The real-time visual analysis method for smart door locks based on edge computing according to claim 3 is characterized in that: The method for constructing the registered user feature database includes: During the user registration phase, millimeter-wave radar is used to collect multiple sets of dynamic frequency shift baseline data, including natural head movements, blinking, and breathing-related signals; After denoising the original frequency shift data, the frequency shift amplitude mean and phase change variance are extracted as dynamic baseline features; The dynamic baseline features are bound to the three-dimensional face template and stored in the trusted execution environment of the edge storage device after encryption.

5. The real-time visual analysis method for smart door locks based on edge computing according to claim 4 is characterized in that: The dual verification of the comprehensive static feature similarity and dynamic frequency shift trajectory consistency specifically includes: The similarity between the three-dimensional topological structure and the template exceeds a first threshold; The deviation of the Doppler frequency motion trajectory from the baseline is lower than a second threshold; If both of the above conditions are met, the verification is successful.

6. The real-time visual analysis method for smart door locks based on edge computing according to claim 5 is characterized in that: The calculation steps of the deviation between the Doppler frequency dynamic trajectory and the baseline are: Perform dynamic time warping alignment on the real-time frequency shift trajectory and the baseline data, and calculate the minimum path distance between the trajectories; If the path distance exceeds the second threshold, it is determined to be a fake motion pattern.

7. The real-time visual analysis method for smart door locks based on edge computing according to claim 5 is characterized in that: The encrypted upload of abnormal features specifically includes: The edge computing node desensitizes the composite feature vectors that fail to match, retaining the key parameters of the frequency shift trajectory and the hash value of the three-dimensional topology structure; Use the door lock device key to homomorphically encrypt the desensitized data, generate verifiable ciphertext and upload it to the cloud threat analysis platform.

8. The real-time visual analysis method for smart door locks based on edge computing according to claim 5 is characterized in that: The verification further includes verifying the confidence level of liveness detection, and the verification of the confidence level of liveness detection includes: Perform spectrum analysis on the Doppler frequency shift trajectory to verify whether it conforms to the movement laws of living organisms.

9. A real-time visual analysis system for smart door locks based on edge computing, used to implement the real-time visual analysis method for smart door locks based on edge computing according to any one of claims 1 to 8, characterized in that: include: A multimodal data acquisition module, integrated into the smart door lock, includes a camera and millimeter-wave radar. It is used to capture the user's facial image in real time, simultaneously obtain the distance between the face and the camera, and extract the Doppler shift characteristics of dynamic motion information based on the millimeter-wave radar echo signal. The edge computing trigger module is deployed on the edge computing node and includes a distance analysis unit, a frequency shift analysis unit, and a liveness determination unit. The distance analysis unit calculates the precise distance value based on the millimeter-wave radar phase difference. The frequency shift analysis unit extracts the periodic motion characteristics of facial micro-movements through time-frequency analysis. The liveness determination unit activates the liveness detection process when it detects that the distance value is within a set range and the multi-shift characteristics match the liveness biometric characteristics. A local feature processing module includes an image feature extraction unit, a depth information registration unit, and a dynamic trajectory encoding unit. The image feature extraction unit generates a two-dimensional texture feature map through a convolutional neural network, the depth information registration unit constructs a three-dimensional topological structure of the face, and the dynamic trajectory encoding unit converts Doppler frequency shift signals into time series features. The spatiotemporal fusion module uses a lightweight model and attention mechanism to perform cross-modal weighted fusion of 2D texture feature maps, 3D topological structures, and time series features to generate a composite living feature vector that includes 3D structure, texture details, and motion characteristics. A dual verification module, connected to the encrypted user feature library in the edge storage device, includes a static feature matching unit and a dynamic trajectory verification unit. The static feature matching unit calculates the similarity between the three-dimensional topology structure and the registration template, and the dynamic trajectory verification unit aligns the real-time frequency shift trajectory with the baseline data through dynamic time warping and verifies the consistency of the spectrum biological regularity. The security execution module includes a door lock control unit, a local alarm unit and an encryption upload unit. The door lock control unit triggers the door opening action when the double verification is passed. The local alarm unit triggers an alarm when the verification fails or a static forgery attack is detected. The encryption upload unit desensitizes and homomorphically encrypts the abnormal features and then uploads them to the cloud threat analysis platform.

10. A computer-readable storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by a processor, a real-time visual analysis method for smart door locks based on edge computing according to any one of claims 1 to 8 is implemented.

Images