Smart lock unlocking method and device based on facial recognition

Abstract

The present invention relates to a smart lock unlocking method and device based on facial recognition, comprising the following steps: performing multi-angle scanning and imaging of the facial area of ​​a user to be unlocked by a depth camera to obtain a three-dimensional facial model; performing heat map analysis on the three-dimensional facial model to obtain a facial heat distribution map; performing liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result, wherein the liveness detection result includes physiological feature information such as facial temperature gradient and blood flow distribution; performing identity authentication based on the three-dimensional facial model and the liveness detection result to obtain an identity authentication result; controlling the smart lock to execute corresponding unlocking or alarm operations according to the identity authentication result to obtain an access control instruction, thereby solving the technical problem that most facial recognition door lock systems mainly rely on two-dimensional images for identity recognition, are easily attacked by counterfeit means such as photos and videos, and have great security risks.

Description

Technical Field

[0001] The present invention relates to the technical field of smart locks, and in particular to a smart lock unlocking method and device based on facial recognition. Background Art

[0002] With the rapid development of artificial intelligence and biometrics, smart door lock systems based on facial recognition are becoming a key application in the smart home and security sectors. Traditional mechanical keys or combination locks, due to their inconvenience in carrying, vulnerability to loss, and potential for cracking, no longer meet the dual demands of security and convenience in modern society. Therefore, utilizing facial recognition technology for contactless authentication is considered an effective way to enhance the intelligence of door locks.

[0003] However, most existing facial recognition door lock systems rely primarily on two-dimensional images for identification, making them vulnerable to forgeries such as photos and videos, posing significant security risks. Furthermore, conventional cameras perform poorly in low light or complex ambient lighting conditions, impacting system stability and reliability. These technical limitations limit the widespread adoption of facial recognition smart locks in real-world scenarios.

[0004] To address these issues, recent research has begun to incorporate multimodal sensors, such as depth cameras and infrared thermal imagers, to improve recognition accuracy and anti-counterfeiting capabilities. However, effectively integrating data from these different sensors, improving liveness detection accuracy, and achieving fast and reliable identity verification remain challenges. Furthermore, system real-time performance, hardware costs, and power consumption all require further optimization and balancing to ensure widespread adoption of this technology. Summary of the Invention

[0005] The main purpose of this invention is to provide a smart lock unlocking method based on facial recognition, which solves the technical problem that most facial recognition door lock systems mainly rely on two-dimensional images for identity recognition, are easily attacked by forgeries such as photos and videos, and pose a major security risk.

[0006] To achieve the above objectives, the present invention provides a smart lock unlocking method based on facial recognition, which is applied to a smart lock equipped with a depth camera and an infrared thermal imager, and comprises the following steps:

[0007] The depth camera scans the facial area of ​​the user to be unlocked at multiple angles to obtain a three-dimensional facial model;

[0008] performing a heat map analysis on the facial area using an infrared thermal imager to obtain a facial heat distribution map;

[0009] Performing liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result;

[0010] Performing identity authentication based on the three-dimensional facial model and the liveness detection result to obtain an identity authentication result;

[0011] An access control instruction is generated according to the identity authentication result, and based on the access control instruction, the preset smart lock is controlled to perform corresponding unlocking or alarm operations, wherein the access control instruction includes specific execution actions of unlocking permission, unlocking rejection, and triggering an alarm.

[0012] Furthermore, the facial area of ​​the user to be unlocked is scanned and imaged from multiple angles by a depth camera to obtain a three-dimensional facial model, including:

[0013] Performing multi-view structured light projection on the facial area of ​​the user to be unlocked through a depth camera to obtain a facial depth point cloud sequence, and calibrating the spatial coordinates of the facial depth point cloud sequence to obtain a facial depth map;

[0014] Reconstructing a facial surface based on the facial depth map to obtain a facial mesh topology structure, and extracting facial geometric features from the facial mesh topology structure;

[0015] The facial geometric features are three-dimensionally reconstructed by fusing multiple frames of depth images to obtain a three-dimensional facial model.

[0016] Furthermore, the thermal image analysis of the facial area is performed using an infrared thermal imager to obtain a facial heat distribution map, including:

[0017] Performing multi-band infrared response modeling on the surface temperature field in the infrared thermal imager to obtain a thermal radiation intensity distribution matrix, and performing spatial normalization mapping on the thermal radiation intensity distribution matrix to obtain a standardized thermal field distribution map;

[0018] Calculating the local temperature gradient of the facial area based on the standardized thermal field distribution map to obtain a temperature gradient vector, and performing directional consistency correction on the temperature gradient vector to obtain an optimized heat flow direction field;

[0019] Performing regional thermal energy aggregation on the optimized heat flow direction field through dynamic neighborhood weighted fusion to obtain a local hotspot response map, and detecting the hot spot center coordinates in the local hotspot response map at multiple scales;

[0020] Based on the center coordinates of the hot spot, a heat propagation path simulation is performed on the facial area to obtain a heat propagation trajectory cluster, and a density cluster analysis is performed on the heat propagation trajectory cluster to obtain a facial heat distribution map.

[0021] Furthermore, performing liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result includes:

[0022] Performing multi-scale spatial frequency decomposition on the facial heat distribution map to obtain a multi-band thermal field energy distribution map, and performing a multi-directional convolution operation on the multi-band thermal field energy distribution map to obtain a multi-directional thermal characteristic coefficient;

[0023] Based on the facial heat distribution map, identifying microcirculatory heat diffusion paths in the facial area to obtain a circulation thermodynamic characteristic map;

[0024] performing adaptive local binary pattern encoding on the cyclic thermodynamic characteristic map to obtain a thermal texture dynamic encoding matrix;

[0025] Performing dynamic thermal inertia simulation on the facial heat distribution map to obtain a thermal response time series, and performing thermal relaxation curve fitting on the facial region based on the thermal response time series to obtain a thermal relaxation characteristic curve;

[0026] Based on the multi-directional thermal feature coefficients, the thermal texture dynamic coding matrix, and the thermal relaxation feature curve, multimodal physiological feature fusion is performed on the facial area to obtain a liveness confidence score, and a liveness determination threshold comparison is performed based on the liveness confidence score to obtain a liveness detection result.

[0027] Furthermore, the adaptive local binary pattern encoding is performed on the cyclic thermodynamic characteristic map to obtain a thermal texture dynamic encoding matrix, including:

[0028] Performing multi-scale gradient field segmentation on the cyclic thermodynamic characteristic map to obtain a gradient amplitude map and a gradient direction map, and performing non-maximum suppression processing on the gradient amplitude map and the gradient direction map to obtain an edge-preserving gradient map;

[0029] Adaptively setting a window size of the edge-preserving gradient map to obtain a local neighborhood window, and performing multi-resolution grayscale quantization on the cyclic thermodynamic characteristic map based on the local neighborhood window to obtain a quantized grayscale value matrix;

[0030] Performing entropy weight allocation on the local neighborhood window based on the quantized gray value matrix to obtain an adaptive weight coefficient, and performing weighted center-symmetric encoding on the local neighborhood window based on the adaptive weight coefficient to obtain a symmetric binary pattern sequence;

[0031] The quantized gray value matrix is ​​spatially associated mapped based on the symmetric binary pattern sequence to obtain a spatial position association map, and the symmetric binary pattern sequence is sparsely constrained reconstructed based on the spatial position association map to obtain a thermal texture dynamic coding matrix.

[0032] Furthermore, performing spatial position correlation mapping on the quantized gray value matrix based on the symmetric binary pattern sequence to obtain a spatial position correlation map includes:

[0033] Performing spatial hash coding on the symmetric binary pattern sequence to obtain a binary pattern hash index, and performing multi-dimensional feature cross calculation on the binary pattern hash index to obtain a spatial correlation feature vector;

[0034] Performing a non-Euclidean distance measurement on the quantized gray value matrix based on the spatial correlation feature vector to obtain a spatial neighborhood relationship matrix, and constructing a graph theory adjacency matrix based on the spatial neighborhood relationship matrix to obtain a characteristic topological adjacency matrix;

[0035] Performing feature node clustering analysis on the feature topology adjacency matrix to obtain a regional clustering feature set, and performing spatial position association mapping based on the regional clustering feature set to obtain a spatial position association graph.

[0036] Furthermore, performing identity authentication based on the three-dimensional facial model and the liveness detection result to obtain an identity authentication result includes:

[0037] Performing key point positioning mapping on the three-dimensional facial model to obtain facial feature point coordinates, and performing local geometric descriptor calculation based on the facial feature point coordinates to obtain a local feature description matrix;

[0038] Performing joint feature embedding on the local feature description matrix and the liveness detection result through feature space transformation technology to obtain a multimodal feature fusion vector, and performing dynamic weighted combination on the multimodal feature fusion vector to obtain an identity feature representation vector;

[0039] Performing similarity measurement calculation on authorized user feature vectors in a pre-stored authorized user feature library based on the identity feature representation vector to obtain an identity matching sequence, and evaluating the credibility of the identity matching sequence to obtain an identity authentication credibility index, wherein the identity authentication credibility index includes a feature matching score, an identity similarity threshold, a verification confidence interval, and an anomaly measurement parameter;

[0040] A multi-threshold joint judgment is performed on the identity authentication credibility index to obtain an identity authentication result.

[0041] The present invention also provides a smart lock unlocking device based on facial recognition, which is applied to a smart lock. The smart lock is provided with a depth camera and an infrared thermal imager, including:

[0042] A scanning module is used to scan the facial area of ​​the user to be unlocked from multiple angles using a depth camera to obtain a three-dimensional facial model;

[0043] an analysis module, configured to perform a heat map analysis on the facial area using an infrared thermal imager to obtain a facial heat distribution map;

[0044] a detection module, configured to perform liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result;

[0045] A determination module, configured to perform identity authentication determination based on the three-dimensional facial model and the liveness detection result to obtain an identity authentication result;

[0046] The control module is used to generate an access control instruction according to the identity authentication result, and control the preset smart lock to perform corresponding unlocking or alarm operations based on the access control instruction, wherein the access control instruction includes specific execution actions of unlocking permission, unlocking rejection, and triggering an alarm.

[0047] The present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of any one of the above methods when executing the computer program.

[0048] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the steps of any of the above methods are implemented.

[0049] The present invention provides a smart lock unlocking method based on facial recognition, comprising the following steps: performing multi-angle scanning and imaging of the facial area of ​​the user to be unlocked by a depth camera to obtain a three-dimensional facial model; performing heat map analysis on the three-dimensional facial model to obtain a facial heat distribution map; performing liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result, wherein the liveness detection result includes physiological feature information such as facial temperature gradient and blood flow distribution; performing identity authentication judgment based on the three-dimensional facial model and the liveness detection result to obtain an identity authentication result; controlling the smart lock to execute corresponding unlocking or alarm operations according to the identity authentication result to obtain an access control instruction, which solves the technical problem that most facial recognition door lock systems mainly rely on two-dimensional images for identity recognition, are easily attacked by counterfeit means such as photos and videos, and have great security risks. The method achieves the technical effect of being able to complete the identity authentication process in a shorter time in combination with an efficient image processing algorithm while ensuring high security, reducing user waiting time, and improving ease of use and interactive experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 1 is a schematic diagram of the steps of a smart lock unlocking method based on facial recognition in one embodiment of the present invention;

[0051] Figure 2This is a structural block diagram of a smart lock unlocking system based on facial recognition in one embodiment of the present invention;

[0052] Figure 3 It is a schematic block diagram of the structure of a computer device according to an embodiment of the present invention.

[0053] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0055] like Figure 1 As shown, Figure 1 A method for unlocking a smart lock based on facial recognition in one embodiment of the present invention is applied to a smart lock equipped with a depth camera and an infrared thermal imager, and includes the following steps:

[0056] Step S1: Use a depth camera to perform multi-angle scanning and imaging of the facial area of ​​the user to be unlocked to obtain a three-dimensional facial model.

[0057] Specifically, a depth camera scans the user's facial area at multiple angles to create a 3D facial model. This process relies on the depth camera's 3D imaging capabilities. Its core principle is to scan the user's face from multiple angles to construct a 3D facial model with spatial geometric information. When a user stands within the smart lock's recognition range, the depth camera initiates a scanning process and uses techniques such as infrared structured light or time-of-flight (TOF) to obtain distance information from various facial parts. This generates a data point cloud containing 3D spatial features such as facial contours, nose bridge height, and eye socket depth. The system then integrates and optimizes this data using a 3D reconstruction algorithm to form a complete and accurate 3D facial model. For example, in a home door application, the user does not need to perform specific movements or wear any auxiliary equipment; they simply face the smart lock naturally. The depth camera then scans the user from multiple angles, ensuring accurate extraction of key facial features even with slight head tilt or occlusion, providing high-precision basic data support for subsequent identity verification.

[0058] Step S2: performing heat map analysis on the facial area using an infrared thermal imager to obtain a facial heat distribution map.

[0059] Specifically, an infrared thermal imager performs heat map analysis on the facial area to obtain a facial heat distribution map. This process is achieved through non-contact sensing and data collection of the human facial temperature field using infrared thermal imaging technology. An infrared thermal imager can capture infrared radiation energy emitted by the human body and convert it into corresponding temperature values, thereby generating a heat distribution image that reflects the temperature differences between different facial regions. In this solution, after the depth camera completes the construction of a three-dimensional facial model, the infrared thermal imager simultaneously performs a thermal energy scan on the same user's facial area, obtaining thermal radiation information from key areas such as the forehead, nose tip, eye area, and lips. Based on this information, a facial heat map with spatial distribution characteristics is drawn. Through further analysis of this heat map, the system can identify whether it conforms to the expected thermal distribution patterns of a real human face. For example, the local temperature differences caused by the distribution of microvessels on the skin surface of a normal face are not dynamic thermal characteristics of faces displayed in photos or on screens. Taking the application scenario of a home entrance door as an example, at night or in a low-light environment, when a user approaches a smart lock, the infrared thermal imager is not restricted by lighting conditions and can still stably obtain the user's facial heat distribution map, thereby providing a reliable basis for subsequent liveness detection and effectively improving the security and accuracy of identity authentication.

[0060] Step S3: performing liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result.

[0061] Specifically, based on the facial heat distribution map, liveness detection is performed on the facial area to obtain a liveness detection result. This process is achieved by analyzing the temperature differences in different areas of the face and their dynamic change characteristics. First, based on the facial heat distribution map provided by the infrared thermal imager, the system can identify the unique thermal pattern of a real human face. This is because real facial skin will show subtle but specific temperature changes caused by blood circulation, and these features are difficult to replicate in a fake face or a two-dimensional image. Next, the system compares these temperature features with a preset real living face model, and uses an algorithm to determine whether the input facial heat distribution conforms to the thermal characteristics of a living body. For example, in the application scenario of a smart door lock, when a user tries to unlock the door through facial recognition, the system not only relies on the consistency of the facial geometry, but also performs in-depth verification based on the facial heat distribution map. If the detected heat distribution shows an appropriate warm area around the eyes, a slightly cooler area at the tip of the nose, and a moderate temperature difference between the forehead and cheeks, it is inferred that a person is alive and access is granted. Conversely, if the detected heat map shows unnatural temperature consistency or does not conform to the expected human thermal characteristics, the person is judged as non-living and the access request is denied, ensuring improved security. The entire process seamlessly combines the advantages of thermal imaging and biometrics, making identity verification both fast and accurate.

[0062] Step S4: performing identity authentication based on the three-dimensional facial model and the liveness detection result to obtain an identity authentication result.

[0063] Specifically, an identity verification determination is performed based on the three-dimensional facial model and the liveness detection result to obtain an identity verification result. This step achieves a more accurate and secure identity recognition process by fusing and analyzing two independent but complementary biometric features. Specifically, after obtaining the three-dimensional facial model of the user, the system first extracts its geometric features, such as facial contours, facial feature spacing, nose bridge height, and other spatial structural information that are unique to each individual, and compares them with pre-stored three-dimensional facial data of legitimate users. Simultaneously, the liveness detection result, derived from the facial heat map, is combined to determine whether the currently input face represents an individual with real vital signs. Only when the three-dimensional model matches successfully and the liveness detection result confirms "live" does the system determine that the user is legitimate and generate the corresponding identity verification result. For example, in a home door application scenario, when a user stands in front of a smart lock, after scanning and thermal imaging are completed, the system synchronously processes these two types of data in the background. If the user is identified as highly consistent with the registered model and possesses the expected thermal distribution characteristics of the human body, identity verification is immediately passed and the unlocking operation is executed, effectively preventing forgery attacks and improving the security and reliability of the entire system.

[0064] Step S5, generating an access control instruction according to the identity authentication result, and controlling the preset smart lock to perform corresponding unlocking or alarming operations based on the access control instruction, wherein the access control instruction includes specific execution actions of unlocking permission, unlocking rejection, and triggering an alarm.

[0065] Specifically, an access control instruction is generated based on the identity authentication result, and based on the access control instruction, the preset smart lock is controlled to perform the corresponding unlocking or alarm operation. The access control instruction includes specific execution actions such as unlocking permission, unlocking rejection, and triggering an alarm. This step is a key link for the entire system to achieve secure access control. After completing the identity authentication based on the facial three-dimensional model and the liveness detection result, the system will generate the corresponding access control instruction based on the verification result. For example, when the identity authentication result confirms that the current user is a legitimate user, the system will generate an "unlocking permission" control instruction, which will transmit the signal to the drive device of the smart lock through the communication module, thereby starting the motor to unlock the door lock. If the identity authentication fails, an "unlocking rejection" instruction will be generated to prevent the door lock from opening. When an abnormal situation is detected, such as multiple consecutive authentication failures or a counterfeit attack is identified, the system will further generate a "triggering alarm" instruction to activate the alarm device to alert the user or security personnel to handle it. Taking the home entrance application scenario as an example, once the user completes facial recognition and successfully passes identity verification, the smart lock will automatically open within seconds. However, if the system recognizes that the currently entered face does not match the registered information or determines that it is a photo attack, it will not only not unlock the door but also trigger an audible and visual alarm, effectively ensuring home security. This process realizes a complete closed-loop management from biometric recognition to physical access control.

[0066] In a specific embodiment, the method of performing multi-angle scanning and imaging of the facial area of ​​the user to be unlocked by a depth camera to obtain a three-dimensional facial model includes:

[0067] Performing multi-view structured light projection on the facial area of ​​the user to be unlocked through a depth camera to obtain a facial depth point cloud sequence, and calibrating the spatial coordinates of the facial depth point cloud sequence to obtain a facial depth map;

[0068] Reconstructing a facial surface based on the facial depth map to obtain a facial mesh topology structure, and extracting facial geometric features from the facial mesh topology structure;

[0069] The facial geometric features are three-dimensionally reconstructed by fusing multiple frames of depth images to obtain a three-dimensional facial model.

[0070] Specifically, the depth camera scans the user's facial area from multiple angles to generate a 3D facial model. This process involves three key steps: first, the depth camera projects multi-view structured light onto the user's facial area to generate a facial depth point cloud sequence, which is then calibrated to create a facial depth map. Second, the facial surface is reconstructed based on the facial depth map to generate a facial mesh topology, and facial geometric features are extracted from the mesh topology. Finally, the facial geometric features are fused into a 3D reconstruction to generate a 3D facial model. This entire implementation process leverages the depth camera's high-precision spatial perception capabilities, combined with structured light technology and 3D reconstruction algorithms, to construct a user's facial model with rich geometric details and accurate spatial position information. In practice, when a user stands in the smart lock's recognition area, the depth camera projects multiple structured light patterns (such as stripes or dots) onto their face. By capturing the deformation of these light patterns across different facial regions, the camera calculates the depth value of each pixel, thereby generating a series of facial depth point cloud data. The system then uses calibration parameters to spatially align these point cloud data to a common reference coordinate system, generating an accurate facial depth map. This depth map not only reflects the contours of the user's facial features but also incorporates individual 3D geometric information, such as nose bridge height and eye socket depth. Based on this depth map, the system uses surface reconstruction algorithms (such as Poisson reconstruction or moving least squares) to construct a continuous and smooth facial mesh topology. It then extracts key facial geometric features for identification, such as the spatial distances between facial features and changes in facial curvature. To enhance the model's integrity and robustness, the system also fuses depth images captured at multiple times to eliminate missing areas caused by slight head movement or occlusion, ultimately generating a complete, highly accurate 3D facial model. For example, when a user approaches a smart lock to open the door, the depth camera immediately initiates the scanning process, completing the entire process from structured light projection, point cloud acquisition, depth mapping, and 3D reconstruction within seconds. Even if the user's face is tilted or partially obscured (e.g., wearing glasses), the system can still effectively restore the complete facial structure through multi-view scanning and multi-frame fusion technology, providing high-quality data support for subsequent identity verification. This 3D modeling method based on the fusion of structured light and depth images not only improves recognition accuracy, but also significantly enhances the system's stability and anti-interference capabilities in complex environments, fully meeting the dual security and real-time requirements of smart locks.

[0071] In a specific embodiment, the heat map analysis of the facial area using an infrared thermal imager to obtain a facial heat distribution map includes:

[0072] Performing multi-band infrared response modeling on the surface temperature field in the infrared thermal imager to obtain a thermal radiation intensity distribution matrix, and performing spatial normalization mapping on the thermal radiation intensity distribution matrix to obtain a standardized thermal field distribution map;

[0073] Calculating the local temperature gradient of the facial area based on the standardized thermal field distribution map to obtain a temperature gradient vector, and performing directional consistency correction on the temperature gradient vector to obtain an optimized heat flow direction field;

[0074] Performing regional thermal energy aggregation on the optimized heat flow direction field through dynamic neighborhood weighted fusion to obtain a local hotspot response map, and detecting the hot spot center coordinates in the local hotspot response map at multiple scales;

[0075] Based on the center coordinates of the hot spot, a heat propagation path simulation is performed on the facial area to obtain a heat propagation trajectory cluster, and a density cluster analysis is performed on the heat propagation trajectory cluster to obtain a facial heat distribution map.

[0076] Specifically, when analyzing the facial region using an infrared thermal imager to obtain a thermal distribution map, the infrared thermal imager first performs multi-band infrared response modeling of the surface temperature field to obtain a thermal radiation intensity distribution matrix. This involves capturing infrared radiation data at different wavelengths using the infrared thermal imager and constructing a thermal radiation intensity distribution matrix based on this data, which reflects the surface temperature field of the object. This thermal radiation intensity distribution matrix is ​​then spatially normalized to eliminate the effects of non-uniformity caused by the measurement environment or the instrument itself, ensuring that the resulting standardized thermal field distribution map accurately reflects the actual temperature differences across the face. Based on this standardized thermal field distribution map, the local temperature gradient of the facial region can be calculated, thereby obtaining a temperature gradient vector. This process involves accurately calculating the temperature change rate between each pixel in the standardized thermal field distribution map and its neighboring pixels. The resulting temperature gradient vector not only reflects the direction of temperature change but also the rate of change. To improve the accuracy of subsequent processing, these temperature gradient vectors must be calibrated for directional consistency. This step aims to correct for directional deviations caused by noise or other factors, thereby obtaining an optimized heat flow direction field. The optimized heat flow direction field provides key information for understanding facial heat flow and helps identify the primary heat conduction pathways. Next, the optimized heat flow direction field is subjected to regional thermal energy aggregation using dynamic neighborhood weighted fusion technology to generate a local hotspot response map. This process considers the neighborhood size and weight distribution at different locations, highlighting areas with high thermal energy concentration characteristics and forming a clear local hotspot response map. Subsequently, a multi-scale detection algorithm is used to determine the center coordinates of hot spots within the local hotspot response map. This method can locate hot spots at different scales, ensuring that even small but significant temperature changes are not overlooked. Based on the identified hot spot center coordinates, the heat propagation pathways within the facial region are simulated, resulting in clusters of heat propagation trajectories. This heat propagation path simulation, based on heat conduction theory and observed temperature distribution, infers how heat diffuses from high-temperature to low-temperature areas. Finally, density cluster analysis is performed on the heat propagation trajectory clusters to extract the facial heat distribution map. Density cluster analysis is a method that identifies high-density regions in a dataset. In this scenario, it is used to identify regions with similar heat propagation characteristics and group them into clusters. The goal is to better understand the thermal relationships between different facial components and their thermal response characteristics under different conditions. For example, in the smart security field, when a user approaches an access control system equipped with an infrared thermal imaging camera, the system can quickly and accurately obtain a thermal map of the user's face using the above steps. This unique thermal signature enables accurate authentication, even in low-light conditions or in the face of disguise attempts.This in-depth analysis method based on infrared thermal imaging not only improves the security and reliability of identification, but also broadens the application scope of biometric technology.

[0077] In a specific embodiment, performing liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result includes:

[0078] Performing multi-scale spatial frequency decomposition on the facial heat distribution map to obtain a multi-band thermal field energy distribution map, and performing a multi-directional convolution operation on the multi-band thermal field energy distribution map to obtain a multi-directional thermal characteristic coefficient;

[0079] Based on the facial heat distribution map, identifying microcirculatory heat diffusion paths in the facial area to obtain a circulation thermodynamic characteristic map;

[0080] performing adaptive local binary pattern encoding on the cyclic thermodynamic characteristic map to obtain a thermal texture dynamic encoding matrix;

[0081] Performing dynamic thermal inertia simulation on the facial heat distribution map to obtain a thermal response time series, and performing thermal relaxation curve fitting on the facial region based on the thermal response time series to obtain a thermal relaxation characteristic curve;

[0082] Based on the multi-directional thermal feature coefficients, the thermal texture dynamic coding matrix, and the thermal relaxation feature curve, multimodal physiological feature fusion is performed on the facial area to obtain a liveness confidence score, and a liveness determination threshold comparison is performed based on the liveness confidence score to obtain a liveness detection result.

[0083] Specifically, liveness detection of the facial region based on the facial heat distribution map and obtaining a liveness detection result is achieved through a series of complex thermal feature extraction and multimodal physiological information fusion techniques. This process first requires multi-scale spatial-frequency decomposition of the facial heat distribution map to obtain the thermal field energy distribution at different frequency levels. This operation utilizes wavelet transform or multi-scale filtering techniques in image processing to decompose the facial surface temperature variation information contained in the original heat distribution map into multiple frequency bands, thereby forming a multi-band thermal field energy distribution map. These frequency bands correspond to different spatial characteristics of facial surface temperature variation. For example, high-frequency components may reflect localized subtle temperature differences, while low-frequency components reflect overall temperature trends. The system then performs a multi-directional convolution operation on the multi-band thermal field energy distribution map to extract the thermal energy variation patterns of the face in different directions and generate multi-directional thermal signature coefficients. This step helps identify the directional characteristics of heat propagation in facial regions, such as temperature gradients along paths from the nose to the corners of the mouth or from the forehead to the eye area, thereby enhancing the perception of facial authenticity. Based on the facial thermal distribution map, the system also identifies microcirculatory heat diffusion pathways within the facial region and generates a circulatory thermodynamic signature map. These microcirculatory pathways are typically associated with the capillary network of the human face, and their heat conduction characteristics exhibit significant dynamics and periodicity, making them an important indicator for determining whether an individual is alive or dead. Furthermore, to capture the dynamic characteristics of facial thermal texture, the system performs adaptive local binary pattern encoding on the circulatory thermodynamic signature map to generate a dynamic thermal texture encoding matrix. This encoding method binarizes temperature contrast relationships within a local region to construct a highly discriminative thermal texture descriptor, effectively representing and quantifying even subtle temperature fluctuations. Furthermore, the system performs dynamic thermal inertia simulation on the facial thermal distribution map to simulate the thermal response of the face to changes in ambient temperature, thereby obtaining a thermal response time series. Based on this time series, the system further performs thermal relaxation curve fitting on the facial region to extract a thermal relaxation signature curve reflecting the thermal properties of the facial material. This curve reveals how quickly and how the face recovers to a stable state after being disturbed, and is a key physical indicator for determining whether it is a human face. Finally, the system integrates the three types of information: the multi-directional thermal feature coefficient, the thermal texture dynamic encoding matrix, and the thermal relaxation characteristic curve, performs a multimodal physiological feature fusion analysis, and calculates a liveness confidence score that reflects whether the current face has liveness attributes. This score integrates information from multiple dimensions such as the face's geometric thermal distribution, dynamic changes in texture, and thermal response behavior, and has a high discriminant ability. The system then compares this score with the preset liveness determination threshold. If the score is higher than the set threshold, it is determined to be "live", otherwise it is determined to be "not alive", thus completing the entire liveness detection process and outputting the liveness detection result.Taking the home entrance application scenario as an example, when a user approaches the smart lock, the infrared thermal imager collects the user's facial heat distribution map in real time, and the system quickly completes the entire process from multi-scale decomposition to multimodal fusion according to the above process. Even if the user attempts to deceive the system by using printed photos or video playback, due to the lack of real microcirculation heat diffusion paths, abnormal thermal texture dynamic encoding, and thermal relaxation curves that do not conform to the thermal inertia of biological tissue, the system can still accurately identify the forgery and deny access requests. This liveness detection mechanism based on deep thermal analysis greatly improves the security and robustness of the smart door lock system, ensuring that only authentic and legitimate users can successfully unlock the door.

[0084] In a specific embodiment, the adaptive local binary pattern encoding of the cyclic thermodynamic characteristic map to obtain a thermal texture dynamic encoding matrix includes:

[0085] Performing multi-scale gradient field segmentation on the cyclic thermodynamic characteristic map to obtain a gradient amplitude map and a gradient direction map, and performing non-maximum suppression processing on the gradient amplitude map and the gradient direction map to obtain an edge-preserving gradient map;

[0086] Adaptively setting a window size of the edge-preserving gradient map to obtain a local neighborhood window, and performing multi-resolution grayscale quantization on the cyclic thermodynamic characteristic map based on the local neighborhood window to obtain a quantized grayscale value matrix;

[0087] Performing entropy weight allocation on the local neighborhood window based on the quantized gray value matrix to obtain an adaptive weight coefficient, and performing weighted center-symmetric encoding on the local neighborhood window based on the adaptive weight coefficient to obtain a symmetric binary pattern sequence;

[0088] The quantized gray value matrix is ​​spatially associated mapped based on the symmetric binary pattern sequence to obtain a spatial position association map, and the symmetric binary pattern sequence is sparsely constrained reconstructed based on the spatial position association map to obtain a thermal texture dynamic coding matrix.

[0089] Specifically, the adaptive local binary pattern encoding (LBPE) of the cyclic thermodynamic feature map is performed to generate a dynamic thermal texture encoding matrix. This process is achieved by multi-level extraction and encoding of the fine structure of the facial surface temperature distribution. First, the system performs multi-scale gradient field segmentation on the cyclic thermodynamic feature map to extract the temperature variation trends of the facial region at different spatial resolutions. This step calculates the gradient magnitude and direction of the image to generate a gradient magnitude map and a gradient direction map, respectively. The gradient magnitude map reflects the severity of temperature variations at each facial point, while the gradient direction map indicates the direction of these variations. Subsequently, to preserve key edge information and remove noise interference, the system performs non-maximum suppression on these two maps to generate an edge-preserving gradient map. This map accurately depicts the boundaries of local temperature differences on the face, providing a structural foundation for subsequent encoding. Next, the system adaptively sets the window size based on the edge-preserving gradient map to form a local neighborhood window. "Adaptive" here means that the window size is not fixed but dynamically adjusted based on the gradient strength and complexity of the current image region, allowing for larger windows in areas with gentle temperature variations and smaller windows in areas rich in detail, thus balancing overall efficiency and local accuracy. Based on this local neighborhood window, the system further performs a multi-resolution grayscale quantization on the original cyclic thermodynamic feature map, mapping the temperature value of each pixel to a discrete set of grayscale levels, thereby generating a quantized grayscale value matrix. This process not only simplifies data representation but also provides a unified numerical basis for subsequent encoding. Furthermore, the system assigns entropy weights to each pixel within the local neighborhood window based on the quantized grayscale value matrix. Specifically, the system assigns weights based on the information content of their grayscale distribution, thereby obtaining adaptive weight coefficients. This weighting mechanism emphasizes the importance of regions with high information entropy (i.e., richer variation and more discriminative significance) in encoding. The system then uses these weight coefficients to perform weighted centrosymmetric encoding on the local neighborhood window. This involves converting the grayscale relationship between the center point of the window and its surrounding pixels into a symmetric binary sequence, forming a symmetric binary pattern sequence. This encoding method not only preserves the contrast relationship of the local thermal texture but also enhances robustness to changes in lighting or environmental conditions. Next, the system maps the spatial correlation of the quantized grayscale value matrix based on the symmetric binary pattern sequence, constructing a spatial correlation map to record the spatial correlation between different regions in thermal texture. Finally, combined with this spatial correlation map, the system performs sparse constrained reconstruction on the symmetric binary pattern sequence to remove redundant information and enhance encoding compactness, ultimately generating a thermal texture dynamic encoding matrix. This matrix not only captures the thermal texture features of the facial region at different scales but also effectively reflects its dynamic changes, providing strong support for subsequent liveness detection.Taking the application scenario of a home entrance door as an example, when the user approaches the smart lock to unlock it, the infrared thermal imager collects the heat distribution information of his or her face, and the system completes the entire process from the cyclic thermodynamic characteristic map to the thermal texture dynamic encoding matrix in sequence. Even at night or in low-light environments, the system can accurately capture the tiny temperature differences and their changing patterns of the face through the above-mentioned adaptive encoding method, thereby identifying whether it is a real face. For example, although a printed photo may be visually close to the appearance of a human face, its thermal texture does not have real dynamic changes, and it is impossible to form an effective thermal texture dynamic encoding matrix through entropy weight allocation and sparse reconstruction. Therefore, it will be identified as a forged object by the system and access will be denied. This authentication mechanism based on dynamic modeling of thermal textures greatly improves the security and reliability of smart locks and ensures the authenticity and stability of identity recognition.

[0090] In a specific embodiment, performing spatial position correlation mapping on the quantized gray value matrix based on the symmetric binary pattern sequence to obtain a spatial position correlation map includes:

[0091] Performing spatial hash coding on the symmetric binary pattern sequence to obtain a binary pattern hash index, and performing multi-dimensional feature cross calculation on the binary pattern hash index to obtain a spatial correlation feature vector;

[0092] Performing a non-Euclidean distance measurement on the quantized gray value matrix based on the spatial correlation feature vector to obtain a spatial neighborhood relationship matrix, and constructing a graph theory adjacency matrix based on the spatial neighborhood relationship matrix to obtain a characteristic topological adjacency matrix;

[0093] Performing feature node clustering analysis on the feature topology adjacency matrix to obtain a regional clustering feature set, and performing spatial position association mapping based on the regional clustering feature set to obtain a spatial position association graph.

[0094] Specifically, when mapping the spatial correlation of a quantized grayscale value matrix based on a symmetric binary pattern sequence to obtain a spatial correlation map, the system first performs spatial hash coding on the symmetric binary pattern sequence, generating binary pattern hash indices. Specifically, spatial hash coding is a technique that converts high-dimensional data into low-dimensional hash codes. Here, it is used to compress complex symmetric binary pattern sequences into more easily processed binary pattern hash indices. The system then uses these hash indices to perform multidimensional feature cross-calculation, analyzing the interactions between features across different dimensions to extract deep spatial correlation feature vectors. This computational approach not only captures the direct relationships between pixels in the original image but also reveals more complex and indirect spatial connections between them. Next, the system applies a non-Euclidean distance metric to the quantized grayscale value matrix based on the spatial correlation feature vectors obtained above. Unlike traditional Euclidean distance, this method more accurately measures the similarities or differences between samples while taking into account the inherent structure of the data. In this way, the system can obtain the spatial relationships between each pixel and its neighbors and construct a spatial neighborhood relationship matrix. Based on this matrix, the system then applies graph theory to construct an adjacency matrix, forming a feature topological adjacency matrix. This matrix not only reflects the connectivity and adjacency relationships between individual pixels but also reflects the internal topological structure of the entire image region, providing a basis for further cluster analysis. After obtaining the feature topological adjacency matrix, the system performs feature node clustering analysis to identify clusters of regions with similar thermal texture characteristics. During the clustering process, the algorithm identifies groups of nodes that are closely connected in the feature space and share similar properties, thereby forming a set of regional cluster features. This step is crucial for understanding local consistency and global distribution in the image, as it helps distinguish different temperature variation patterns or thermal texture types. Finally, based on this set of regional cluster features, the system performs spatial position correlation mapping, ultimately generating a spatial position correlation graph. This graph not only displays the relative positional relationships between different regions but also reveals potential connections between them in terms of thermodynamic characteristics, providing an intuitive visualization tool for subsequent analysis. For example, in a smart door lock application scenario, when a user attempts to unlock the door, the system can collect a cyclic thermodynamic feature map of their face using an infrared sensor and process it sequentially according to the aforementioned steps. When faced with attempts at night or in low-light environments, the system accurately distinguishes real faces from forged photos or other forms of deception. Even though a photo may superficially mimic a face's appearance, it lacks the true temperature distribution and dynamic characteristics of a real face and cannot produce a spatial correlation map that matches a real person. Therefore, the system relies on the spatial correlation map to verify the user's legitimacy, ensuring that only genuine users can successfully unlock the device.In addition, since this method combines multi-level spatial information processing techniques, including the initial symmetric binary pattern encoding and the final spatial position association mapping, it exhibits higher robustness and accuracy in the face of various complex environments and challenges.

[0095] In a specific embodiment, performing identity authentication based on the three-dimensional facial model and the liveness detection result to obtain the identity authentication result includes:

[0096] Performing key point positioning mapping on the three-dimensional facial model to obtain facial feature point coordinates, and performing local geometric descriptor calculation based on the facial feature point coordinates to obtain a local feature description matrix;

[0097] Performing joint feature embedding on the local feature description matrix and the liveness detection result through feature space transformation technology to obtain a multimodal feature fusion vector, and performing dynamic weighted combination on the multimodal feature fusion vector to obtain an identity feature representation vector;

[0098] Performing similarity measurement calculation on authorized user feature vectors in a pre-stored authorized user feature library based on the identity feature representation vector to obtain an identity matching sequence, and evaluating the credibility of the identity matching sequence to obtain an identity authentication credibility index, wherein the identity authentication credibility index includes a feature matching score, an identity similarity threshold, a verification confidence interval, and an anomaly measurement parameter;

[0099] A multi-threshold joint judgment is performed on the identity authentication credibility index to obtain an identity authentication result.

[0100] Specifically, the identity verification process based on a 3D facial model and liveness detection results involves a multi-layered, multimodal data processing and analysis process designed to ensure both accuracy and security. First, the system performs key point mapping on the 3D facial model. This involves identifying key facial features, such as the eyes, nose, and mouth, and obtaining their precise coordinates in 3D space. Using these facial feature point coordinates, the system calculates local geometric descriptors (LGDs), mathematical representations of the shape characteristics of specific regions, resulting in a local feature description matrix that captures facial structural details. Next, to fully leverage information from diverse data sources, the system employs feature space transformation techniques to perform joint feature embedding on the LDFD matrix and liveness detection results, generating a multimodal feature fusion vector. This vector integrates facial structural information and biometric information confirming the user's identity, enhancing the reliability and anti-spoofing capabilities of identity verification. Then, by dynamically weighting this multimodal feature fusion vector, adjusting the importance of each feature according to different weighting strategies, the system extracts a more refined identity representation vector. This step is crucial for balancing the contributions of features from different sources, helping to improve the accuracy of the final decision. Based on the obtained identity feature representation vector, the system calculates a similarity metric against the authorized user feature vectors in a pre-stored authorized user feature database, generating an identity match sequence. This sequence reflects the degree of similarity between the current subject and all known authorized users in the database. To assess the confidence of this match, the system calculates a series of metrics, including but not limited to a feature match score (indicating the degree of match between the two sets of features), an identity similarity threshold (a set minimum similarity requirement), a verification confidence interval (a range of uncertainty in the match result), and anomaly metrics (used to detect potential fraud). Combining these metrics, the system generates a comprehensive identity authentication confidence index. Finally, the system performs a multi-threshold joint decision on this identity authentication confidence index, a complex decision-making process based on multiple criteria and conditions. By comparing each metric against pre-set thresholds, the system makes a final authentication decision. If the match is high and no anomalies are found, the system confirms the user's identity. Conversely, if any inconsistencies or suspicions are detected, the access request may be denied or additional security checks may be triggered. The entire process, from identifying key points to final authentication results, is designed to provide a secure, efficient, and reliable personal identity authentication mechanism. For example, in a bank's automated teller machine (ATM) scenario, users can withdraw cash using facial recognition. The system uses this method to ensure that only legitimate users can successfully conduct transactions, effectively preventing unauthorized access.

[0101] The above describes the smart lock unlocking method based on facial recognition in the embodiment of the present invention. The following describes the smart lock unlocking device based on facial recognition in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, a smart lock unlocking device based on facial recognition includes:

[0102] The scanning module 21 is used to scan the facial area of ​​the user to be unlocked from multiple angles using a depth camera to obtain a three-dimensional facial model;

[0103] An analysis module 22 is configured to perform a heat map analysis on the facial area using an infrared thermal imager to obtain a facial heat distribution map;

[0104] A detection module 23 is configured to perform liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result;

[0105] A determination module 24 is configured to perform identity authentication based on the three-dimensional facial model and the liveness detection result to obtain an identity authentication result;

[0106] The control module 25 is used to generate an access control instruction according to the identity authentication result, and control the preset smart lock to perform corresponding unlocking or alarm operations based on the access control instruction, wherein the access control instruction includes specific execution actions of unlocking permission, unlocking rejection, and triggering an alarm.

[0107] In this embodiment, for the specific implementation of each unit in the above device embodiment, please refer to the above method embodiment, which will not be repeated here.

[0108] Reference Figure 3 In an embodiment of the present invention, a computer device is also provided, wherein the internal structure of the computer device can be as follows: Figure 3 As shown. The computer device includes a processor, memory, display screen, input device, network interface and database connected via a system bus. The processor of the computer design is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and computer program in the non-volatile storage medium. The database of the computer device is used to store the corresponding data in this embodiment. The network interface of the computer device is used to communicate with an external terminal via a network connection. When the computer program is executed by the processor, the above method is implemented.

[0109] Those skilled in the art will understand that Figure 3 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention and does not constitute a limitation on the computer device to which the solution of the present invention is applied.

[0110] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which implements the above-described method when executed by a processor. It is understood that the computer-readable storage medium in this embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0111] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing the relevant hardware using a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the above-described method embodiments. Any reference to memory, storage, database, or other media provided herein and used in the embodiments may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double-speed SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct RAMbus dynamic RAM (DRDRAM), and RAMbus dynamic RAM.

[0112] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, apparatus, article, or method comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, apparatus, article, or method. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, apparatus, article, or method comprising the element.

[0113] The above description is only a preferred embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made by using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A smart lock unlocking method based on facial recognition, characterized in that: Applied to a smart lock, the smart lock is equipped with a depth camera and an infrared thermal imager, including the following steps: The depth camera scans the facial area of ​​the user to be unlocked at multiple angles to obtain a three-dimensional facial model; performing a heat map analysis on the facial area using an infrared thermal imager to obtain a facial heat distribution map; Performing liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result; Performing identity authentication based on the three-dimensional facial model and the liveness detection result to obtain an identity authentication result; Generate an access control instruction according to the identity authentication result, and control the preset smart lock to perform corresponding unlocking or alarming operations based on the access control instruction, wherein the access control instruction includes specific execution actions of unlocking permission, unlocking rejection, and triggering an alarm; The thermal image analysis of the facial area by an infrared thermal imager to obtain a facial heat distribution map includes: Performing multi-band infrared response modeling on the surface temperature field in the infrared thermal imager to obtain a thermal radiation intensity distribution matrix, and performing spatial normalization mapping on the thermal radiation intensity distribution matrix to obtain a standardized thermal field distribution map; Calculating the local temperature gradient of the facial area based on the standardized thermal field distribution map to obtain a temperature gradient vector, and performing directional consistency correction on the temperature gradient vector to obtain an optimized heat flow direction field; Performing regional thermal energy aggregation on the optimized heat flow direction field through dynamic neighborhood weighted fusion to obtain a local hotspot response map, and detecting the hot spot center coordinates in the local hotspot response map at multiple scales; Based on the center coordinates of the hot spot, a heat propagation path simulation is performed on the facial area to obtain a heat propagation trajectory cluster, and a density cluster analysis is performed on the heat propagation trajectory cluster to obtain a facial heat distribution map.

2. The method for unlocking a smart lock based on facial recognition according to claim 1, characterized in that: The method of scanning the facial area of ​​the user to be unlocked by the depth camera at multiple angles to obtain a three-dimensional facial model includes: Performing multi-view structured light projection on the facial area of ​​the user to be unlocked through a depth camera to obtain a facial depth point cloud sequence, and calibrating the spatial coordinates of the facial depth point cloud sequence to obtain a facial depth map; Reconstructing a facial surface based on the facial depth map to obtain a facial mesh topology structure, and extracting facial geometric features from the facial mesh topology structure; The facial geometric features are three-dimensionally reconstructed by fusing multiple frames of depth images to obtain a three-dimensional facial model.

3. The method for unlocking a smart lock based on facial recognition according to claim 1, characterized in that: The performing liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result includes: Performing multi-scale spatial frequency decomposition on the facial heat distribution map to obtain a multi-band thermal field energy distribution map, and performing a multi-directional convolution operation on the multi-band thermal field energy distribution map to obtain a multi-directional thermal characteristic coefficient; Based on the facial heat distribution map, identifying microcirculatory heat diffusion paths in the facial area to obtain a circulation thermodynamic characteristic map; performing adaptive local binary pattern encoding on the cyclic thermodynamic characteristic map to obtain a thermal texture dynamic encoding matrix; Performing dynamic thermal inertia simulation on the facial heat distribution map to obtain a thermal response time series, and performing thermal relaxation curve fitting on the facial region based on the thermal response time series to obtain a thermal relaxation characteristic curve; Based on the multi-directional thermal feature coefficients, the thermal texture dynamic coding matrix, and the thermal relaxation feature curve, multimodal physiological feature fusion is performed on the facial area to obtain a liveness confidence score, and a liveness determination threshold comparison is performed based on the liveness confidence score to obtain a liveness detection result.

4. The method for unlocking a smart lock based on facial recognition according to claim 3, characterized in that: The step of performing adaptive local binary pattern encoding on the cyclic thermodynamic characteristic map to obtain a thermal texture dynamic encoding matrix includes: Performing multi-scale gradient field segmentation on the cyclic thermodynamic characteristic map to obtain a gradient amplitude map and a gradient direction map, and performing non-maximum suppression processing on the gradient amplitude map and the gradient direction map to obtain an edge-preserving gradient map; Adaptively setting a window size of the edge-preserving gradient map to obtain a local neighborhood window, and performing multi-resolution grayscale quantization on the cyclic thermodynamic characteristic map based on the local neighborhood window to obtain a quantized grayscale value matrix; Performing entropy weight allocation on the local neighborhood window based on the quantized gray value matrix to obtain an adaptive weight coefficient, and performing weighted center-symmetric encoding on the local neighborhood window based on the adaptive weight coefficient to obtain a symmetric binary pattern sequence; The quantized gray value matrix is ​​spatially associated mapped based on the symmetric binary pattern sequence to obtain a spatial position association map, and the symmetric binary pattern sequence is sparsely constrained reconstructed based on the spatial position association map to obtain a thermal texture dynamic coding matrix.

5. The method for unlocking a smart lock based on facial recognition according to claim 4, characterized in that: The performing spatial position correlation mapping on the quantized gray value matrix based on the symmetric binary pattern sequence to obtain a spatial position correlation map includes: Performing spatial hash coding on the symmetric binary pattern sequence to obtain a binary pattern hash index, and performing multi-dimensional feature cross calculation on the binary pattern hash index to obtain a spatial correlation feature vector; Performing a non-Euclidean distance measurement on the quantized gray value matrix based on the spatial correlation feature vector to obtain a spatial neighborhood relationship matrix, and constructing a graph theory adjacency matrix based on the spatial neighborhood relationship matrix to obtain a characteristic topological adjacency matrix; Performing feature node clustering analysis on the feature topology adjacency matrix to obtain a regional clustering feature set, and performing spatial position association mapping based on the regional clustering feature set to obtain a spatial position association graph.

6. The method for unlocking a smart lock based on facial recognition according to claim 1, characterized in that: The performing identity authentication determination based on the three-dimensional facial model and the liveness detection result to obtain the identity authentication result includes: Performing key point positioning mapping on the three-dimensional facial model to obtain facial feature point coordinates, and performing local geometric descriptor calculation based on the facial feature point coordinates to obtain a local feature description matrix; Performing joint feature embedding on the local feature description matrix and the liveness detection result through feature space transformation technology to obtain a multimodal feature fusion vector, and performing dynamic weighted combination on the multimodal feature fusion vector to obtain an identity feature representation vector; Performing similarity measurement calculation on authorized user feature vectors in a pre-stored authorized user feature library based on the identity feature representation vector to obtain an identity matching sequence, and evaluating the credibility of the identity matching sequence to obtain an identity authentication credibility index, wherein the identity authentication credibility index includes a feature matching score, an identity similarity threshold, a verification confidence interval, and an anomaly measurement parameter; A multi-threshold joint judgment is performed on the identity authentication credibility index to obtain an identity authentication result.

7. A smart lock unlocking device based on facial recognition, characterized in that: Applied to a smart lock, used to execute the smart lock unlocking method based on facial recognition as described in any one of claims 1 to 6, wherein the smart lock is provided with a depth camera and an infrared thermal imager, comprising: A scanning module is used to scan the facial area of ​​the user to be unlocked from multiple angles using a depth camera to obtain a three-dimensional facial model; an analysis module, configured to perform a heat map analysis on the facial area using an infrared thermal imager to obtain a facial heat distribution map; a detection module, configured to perform liveness detection on the facial area based on the facial heat distribution map to obtain a liveness detection result; A determination module, configured to perform identity authentication determination based on the three-dimensional facial model and the liveness detection result to obtain an identity authentication result; A control module is configured to generate an access control instruction based on the identity authentication result, and control a preset smart lock to perform a corresponding unlocking or alarming operation based on the access control instruction, wherein the access control instruction includes specific execution actions such as unlocking permission, unlocking rejection, and triggering an alarm; The thermal image analysis of the facial area by an infrared thermal imager to obtain a facial heat distribution map includes: Performing multi-band infrared response modeling on the surface temperature field in the infrared thermal imager to obtain a thermal radiation intensity distribution matrix, and performing spatial normalization mapping on the thermal radiation intensity distribution matrix to obtain a standardized thermal field distribution map; Calculating the local temperature gradient of the facial area based on the standardized thermal field distribution map to obtain a temperature gradient vector, and performing directional consistency correction on the temperature gradient vector to obtain an optimized heat flow direction field; Performing regional thermal energy aggregation on the optimized heat flow direction field through dynamic neighborhood weighted fusion to obtain a local hotspot response map, and detecting the hot spot center coordinates in the local hotspot response map at multiple scales; Based on the center coordinates of the hot spot, a heat propagation path simulation is performed on the facial area to obtain a heat propagation trajectory cluster, and a density cluster analysis is performed on the heat propagation trajectory cluster to obtain a facial heat distribution map.

8. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images